The embargo has lifted on Nvidia's RTX Spark platform, and the internet is currently doing what it always does: arguing over spreadsheet math. I've done something different with my diary of a first week living with Spark, but I've been busy benchmarking too!

For months, the tech industry (myself included) hyped up this Arm-based Grace CPU and Blackwell RTX GPU combo as the ultimate "MacBook Pro killer." Reviewers (also like me) have spent the past seven days squinting at Geekbench scores to see if Nvidia finally dethroned Apple's M-series silicon.

Let me save you some time: it didn't.

But judging RTX Spark purely by its single-core CPU speeds is like judging a Ferrari by its cup holders. And after a week of stress-testing the Microsoft Surface Laptop Ultra with Spark — running every synthetic and real-world test we can throw at it — it's clear that this is so much more.

Those numbers tell a very different, far more complicated story. Let me explain.

Where to buy RTX Spark laptops

• MS Surface Laptop Ultra: $2,599 @ MS Store
• HP OmniBook X AI 14: $2,999 @ HP
• HP OmniBook Ultra AI 16 OLED: $3,199 @ Best Buy
• Lenovo Pro 9n: $2,999 @ Lenovo
• Lenovo Yoga 9n 2-in-1: $2,999 @ Lenovo
• MSI Prestige N16 Flip AI+ OLED 2-in-1: preorder for $3,299 @ MSI
• Dell XPS 16: $3,799 @ Best Buy
• Asus ProArt P14: $6,499 @ Asus
• Asus ProArt P16: $4,499 @ Asus

CPU reality check: Apple is still king, but other Windows laptops don't stand a chance

So let's start with the first benchmark everybody goes to. If you were hoping Nvidia's Arm-based Grace CPU was going to embarrass the M5 Pro, you need to adjust your expectations.

Given how long it took to make this, it's fair to say this was the main bit I felt nervous about. In short, not terrible, but underwhelming.

That being said, this was one of the first chances I had to test Nvidia's claims of getting the same performance on and off the charger.

Now I know there's a difference maker in here — Windows power settings. Putting it to best efficiency will probably starve the chip of power to match it. With that in mind, I've just done the test in both this mode and best performance to see what I get.

Certainly a surprise to see it gain a few points when unplugged, but this falls within the window of variability in the test results I've been getting. So the answer is a resounding "yes," provided you slap on best performance.

To dig deeper into that multi-core performance, I went to Handbrake. One of the tests we run across all systems is transcoding a copy of Tears of Steel from 4K down to 1080p. This is a solid real-world test of video encoding and real-world multi-core processing.

As you can see, it clearly leads the way in Windows laptops (Intel caught napping there), but Apple silicon reigns as champ.

The GPU flex: Comparable to an RTX 5070 laptop

Gaming is where Nvidia gets its lick back against the competition. It has a massive Blackwell RTX GPU in it, after all, and when you look at the 3DMark numbers, it matches up to something you’d see in an RTX 5070 gaming laptop — even when tasked with x86 emulation in some of these tests.

And to check it's not just the CPU getting that same performance on and off the charger...

Once again, same performance on and off the battery. This is consistent with what Nvidia's been telling us, and it's one of the features gatekept by Apple MacBooks that is now in the hands of Windows users!

Now, let's get into the games.

Forza Horizon 6

Forza Horizon 6 is one of my favorite racing games of the past decade, so it only makes sense I start here. Let's max it without DLSS tech and see what we get.

Bear in mind this is a stress test of epic proportions, but you can see how it struggles. That 1% low is a good indicator of the consistency of the frame rate, which is highly volatile here.

Interestingly, as I tried this push to Extreme + RT settings, I noticed some rubberbanding in the gameplay — probably down to the CPU having to work overtime translating the x86 code of the game while also sending the instructions to the GPU.

You can see that as Frametime spikes can be seen anywhere from 60ms up to 140ms. The game feels sluggish.

But now let's hone it. Thinking about this like an RTX 5070 laptop, I'll drop it down to Ultra + RT (same resolution) and stick with DLSS 4.5 set to auto with dynamic frame generation.

Immediately, this feels so much better to play! There are still some latency spikes here and there, but none that are hardly noticeable.

Also interesting to note, the temperatures on the GPU dropped too by one degree (71 C in the first run, and 70C in the second).

And the cherry on top — here's performance on battery.

Yep, same performance on and off the charger here!

Cyberpunk 2077

Listen up, choombas! It wouldn't be an Nvidia laptop test if we didn't touch on Cyberpunk 2077. For context on this one, I'm gonna compare it against gaming laptops with different GPUs, so you can see where it lands.

And if you're curious, I got 38.9 FPS when unplugged (basically 39).

Black Myth: Wukong

Is this a safe space? ...I'm really not a fan of Black Myth: Wukong.

OK, I get it! I can appreciate the craft of it, and I get why people love it. I've just never got on with it. That being said, I did benchmarks for this too.

...you spotting a pattern? This maxed-out RTX Spark is lining up pretty much neck-and-neck with the RTX 5070, while also running cooler and more efficiently. Nifty!

Gears of War: E-Day

This came out recently, and I saw it being played at the Surface Laptop Ultra event in San Francisco, so of course I had to push it further here.

This is running at 2180x1453 Ultra settings with DLSS 4.5 Dynamic and RTX Mega Geometry turned on.

Some observations here — this is clearly a scalable engine on PC that can run fast here. But there were some micro-stutters when I pushed it to the bleeding edge.

That being said, the average frame time and response to controller inputs meant this all felt smooth with virtually zero latency.

Creative apps is where RTX Spark starts to shine

There are several ways to test creative performance, so I'll dive into them one by one. PugetBench is a good way of seeing how a laptop will fare in your standard creative prosumer workload across Photoshop and Premiere Pro. Here's how RTX Spark fared in my testing.

The finely-tuned cores of the M5 Pro take the win here, but RTX Spark does take a leap ahead of its Windows competition in creator performance.

Let's get to Cinebench

Unfortunately, there isn't any GPU testing available here yet (the OptiX driver for the chip isn't there), so this is a straight CPU comparison.

These numbers put it ahead of Intel, AMD, and Qualcomm's latest flagships, but Apple M5 Pro (again) comes out on top. It's the best of the Windows bunch.

AI performance is where RTX Spark takes the lead

To put some numbers to Nvidia's advantage here, I ran it through UL Procyon AI benchmarks.

This isn't a test that has been normally run on other laptops in the Tom's Guide lab, so I've had to ask around my friends/fellow nerds for insights on this.

In their experience, RTX Spark seems to be surpassing the M5 Pro's AI prowess, exceeding the capabilities of AMD Ryzen AI Max+ and blowing past Intel Core Ultra X9. Once I can do these tests myself, I'll do a follow-up.

Also, for what it's worth, here's the comparison to my RTX 5060 Ti 16GB desktop GPU.

A reduction is always expected between the raw wattage of desktop parts and a laptop chip, but the Spark does get impressively close.

Battery life is a mixed bag

So what about stamina? Nvidia promised all-day battery life, and technically that’s true. But it doesn’t come close to the efficiency king of Apple’s MacBook Pro.

When it comes to gaming or running local AI at full pelt, this falls in line with what you’d expect from an RTX gaming laptop (our gaming test result got one hour and thirty-eight minutes) — give or take about 20-30 minutes depending on how much power the internals soak up.

Meanwhile, if you're expecting that same all-day battery life claim to extend to gaming, that's not the case. You do indeed get more stamina than your average gaming laptop, but not by much.

Who should buy RTX Spark, and who should buy a gaming laptop?

That battery equivalency does raise another question — who is RTX Spark for?

Because for all intents and purposes, an RTX 50-series gaming laptop could do a lot of this workload, whether it’s gaming, creative work, or some of the local AI stuff. And depending on how efficient that gaming laptop can run, you could get a similar battery life.

The real differentiator comes in that unified memory. With there being two pools of memory between the RAM and the GPU’s VRAM in a gaming laptop, it’s not as efficient as having one giant tub of it that can be divided depending on job priority.

That means on RTX Spark, you can run vastly larger local AI models, and larger projects in the likes of Blender than you could elsewhere. This is definitely for the prosumer market.

More from Tom's Guide

• I didn’t believe an Nvidia RTX Spark laptop would have the same performance both on and off the charger — so I put it to the test
• Forget RTX Spark: The Surface Laptop Ultra’s magnetic USB-C charger is a game-changer that makes Apple’s MagSafe look obsolete
• I just tested Windows 11 26H2 — here’s how to update now and why Microsoft finally fixed the two worst things about Windows 11

Let me be clear from the top. If you are looking for the raw benchmark data, the thermal charts and synthetic scores for Nvidia’s RTX Spark, you can head over to my testing lab here.

…ok are all the speeds and feeds people gone? Good, because while I’m over there doing the math, and everyone else on the internet combs over the numbers, data doesn’t explain what it actually feels like to use a machine with a locally hosted AI brain.

You have to unlearn years of cloud-tethered habits (like relying on a browser-based chatbot tab) and start treating your operating system like an overly eager digital intern. And over the past few days, I found this out and quite frankly, it’s a bit of a breakthrough.

So, before you dive into the frame rates, here is a diary of my first week ditching the M5 MacBook Pro, and living exclusively on an Microsoft Surface Laptop Ultra with RTX Spark (and 128GB of unified memory).

Day 1: The Sims 4 disaster (and a lesson in patience)

I returned from the Surface Laptop Ultra event in San Francisco pretty late, but I didn’t feel tired. I was caught up in the excitement of tearing this machine out of the box and seeing what it could do.

And my first test? Nothing to do with productivity. Instead, I installed Hermes agent with Qwen 3.8 Flash Next to play The Sims 4 — build a house and get a job.

Now with that out the way, this first test did not instill me with immediate confidence in Nvidia’s vision of a local agentic future. The AI fumbled around, got lost in the house building menu, and when I got it to the point of having a home roughly three hours later, it managed to start a kitchen fire.

Throughout that time, my prompts to the agent grew more frustrated, and it repeatedly made the same mistakes — forgetting that the graph of co-ordinates keeps moving with the character in the center of the screen and going round in circles.

But watching it stumble taught me a crucial lesson in reinforcement learning. No, not that reinforcement learning that's the complex phase of training an AI model that researchers work hard on. This is more in the human, psychological sense — rewarding good behavior and correcting bad behavior in real-time.

I had another go in the morning with a calmer head, continued to tell it when it was doing the right thing, encouraging it to keep going and we made it to the first gym.

You see, interacting with a local agent isn’t like querying a search engine or talking to a cloud frontier model; it’s exactly like playing two-player Portal 2 with my eight-year-old nephew. You can’t get frustrated when he jumps into the water and dies for the 5th time. You have to give him the carrot, not the stick — guide with positive reinforcement and structured boundaries.

Once that clicked in my mind and my behavior changed, learning how to actually “parent” the agent immediately skyrocketed the output quality. And that’s when I truly got to work.

Day 2: The butler arrives

By the next day, I booted up the Hermes agent and built a set of custom skills. I hooked it directly to my personal Google Workspace API, and for my Tom’s Guide account (Future locks down API access pretty tightly), I set up a script allowing it to physically operate my browser.

I named it Jeeves, and every morning, I just write “check-emails” and it gets to work across both accounts — reading my inboxes, deleting anything unnecessary (like marketing emails or phishing scams), drafts surprisingly human responses to those that are important, and generates a prioritized to-do list for my workday based on emails and Slack.

Now I know what you’re thinking. Isn’t this possible on other machines? Yes. Hermes is an agentic machine that works across Macs and other Windows PCs too (including AMD machines and we all know how they feel about the RTX Spark hype).

But Nvidia’s GPU is the key advantage here, and it comes down to CUDA (which stands for Compute Unified Device Architecture). Basically, it's a technology that unlocked the power of the GPU for other purposes (like AI in this instance).

It’s the reason why Team Green is the biggest picks and shovels company in the cloud AI gold rush (or bubble…however you see it), and an overwhelming majority of open-source models used for local AI (including the Qwen model I’ve been using) are built, trained and initially optimized specifically for Nvidia’s CUDA platform.

There’s a noticeable speed advantage in AI between RTX Spark systems and trying Hermes Agent on my MacBook Pro. On top of that, you're seeing Spark lead substantially against Ryzen AI Max+ 395 (I haven't had access to test 495 PRO).

I’m getting into the weeds here, the difference maker here comes down to my awful work-life balance. I usually sleep with a laptop next to me (...awful, I know), and when the alarm goes off at 8am, I’m right back on my MacBook checking emails and figuring out what to do that day.

Now, I can just throw my agent at it, go make breakfast and coffee and chill out talking to my wife before the day begins at 9am. It’s these small but significant time saves that I feel are the most important.

Day 3: Embracing the creator chaos

Mid-week, I dove into content creation. This is kind of the RTX Spark’s bread and butter, because while the Arm-based Grace CPU has noticeably slower single-core speeds compared to the current heavyweight champions over at Apple, its multicore performance is a runaway freight train.

Throw in that massive Blackwell GPU, and I was having a ball blasting through a project on an open-source creative suite.

Then, I tackled ComfyUI. At first glance, the node-mapping interface is a chaotic, terrifying mess of digital spaghetti. But after a day of tinkering and reading guides, the logic clicks. My first challenge: ask my wife to send me a quick sketch. She doodled a picture of Medusa, which I fed into the local UI.

Within 20 seconds, the RTX Spark chewed through the local rendering and spat out a 90s-style cartoon image of her drawing. Then going a little over the top, I went further with an image-to-video pipeline and made a Saturday morning cartoon-style scene of Medusa turning a man into stone.

The speed differentials of this are massive, and one of the jewels in the RTX Spark's crown — roughly 2-3 seconds faster per image generation in my UL Procyon testing against AMD Ryzen AI Max+, and the MacBook Pro trails too.

I’m no fan of AI slop, but credit where credit’s due. Doing this with a cloud AI model would have cost a premium subscription, and without the finite controls of ComfyUI, it would have taken a lot longer to get the results just right.

Day 4: Mutating code and webcam magic

I’ve got distracted enough by AI. Now it’s time to do the thing that I love — gaming. And being honest, playing on RTX Spark is generally excellent. Turns out the fix to gaming on Arm systems is to just throw a massive GPU at it.

007 First Light runs at 1440p with everything turned up to Ultra, DLSS 4.5 and Dynamic Frame Generation maxes out the refresh rate on this laptop screen, DLSS 5 on NBA 2K27 is still as photorealistic (and divisive, looking at my comments) as ever, and shooters that require low latency rapid reactions run smoothly on here (complete with anti-cheat support).

But there are still some gremlins in the system. Red Dead Redemption 2 hit a brick wall by throwing an incompatibility error and refusing to run.

After speaking to Nvidia, I got a guide to make it run on DirectX 12 rather than Vulkan, which didn’t work through the physical steps given to me. So rather than search Reddit for an answer, I simply asked the Hermes agent to make it work.

The local AI dug into the file director, literally rewrote the offending configuration code to bypass the Arm translation block, and the game booted up. Having your laptop dynamically rewrite the game code to fix its own compatibility issues feels like pure sci-fi.

I also had meetings to go to, so I fired up the new version of Nvidia Broadcast, which is basically DLSS for webcams. This 1080p 30 FPS shooter can embrace that GPU prowess to upscale to 4K60 with a studio light — cleaning up the muddy laptop camera feed in real-time. My team really noticed the difference!

Day 5: The MacBook Pro reckoning

By the end of my first week, I had to face the reality of the silicon and hardware encasing it.

Going into this, everyone — Microsoft (with that $1,000 trade-in offer), tech journalists, and myself included — came in thinking the RTX Spark could be the ultimate MacBook Pro killer. The truth? It’s not.

In my standard browser run-down test, the battery died right around the 10-hour mark. That’s solid for a Windows machine and does fit Nvidia’s claims of all-day battery life, but it isn’t touching Apple’s endurance. Furthermore, Apple’s M5 Pro and Max chips easily outstrip the CPU speeds of this system.

But I realized that comparing it to a MacBook Pro misses the point entirely. The RTX Spark isn’t trying to be a better Mac; it is a different species of computer that has arrived early.

It is a statement of intent from Nvidia about where computing is going — trading raw benchmarking crowns on the stats that we currently see as important for the ability to carry a self-rewriting, fully autonomous, CUDA-accelerated digital brain in your backpack.

It requires a new set of behaviors, but once you let local AI take the wheel in some areas, it makes you believe the future of computing is entirely local (and without a subscription fee).

More from Tom's Guide

• I didn’t believe an Nvidia RTX Spark laptop would have the same performance both on and off the charger — so I put it to the test
• Forget RTX Spark: The Surface Laptop Ultra’s magnetic USB-C charger is a game-changer that makes Apple’s MagSafe look obsolete
• I just tested Windows 11 26H2 — here’s how to update now and why Microsoft finally fixed the two worst things about Windows 11

Last week, AMD fired a warning shot at Nvidia’s RTX Spark. Today, Team Green’s payload has arrived. Ahead of the highly anticipated October 16 launch, Nvidia just unleashed a tidal wave of software for its audacious Arm-based chip, the massive Blackwell RTX GPU.

They aren’t just selling a new chip; they’re trying to prove that their garden of software features makes the jump to Arm completely worth it.

And on top of that, I currently have the Microsoft Surface Laptop Ultra with RTX Spark sitting on my desk (actually typing this story on it right now). Now, before I lock myself in my lab for the review, I want to know exactly what you want to see this thing do.

Here is the cheat sheet of the craziest software witchcraft Nvidia just announced. Read it over, drop your most chaotic testing requests in the Q&A box just below, and I will push this hardware to the absolute limit.

Reminder: We’re a friendly bunch, and this community is a safe space for anyone who wants to learn about RTX Spark. Please treat everyone (including me) with respect and avoid any nasty language.

AI without the computer science degree

Running a Large Language Model (LLM) on your PC used to require opening terrifying command-line terminals and pretending you knew how to code (I’m guilty of that one with my D in Computer Science). That friction is well on its way to being killed, and Nvidia is taking another big step forward.

• Hugging Face Llama Windows: Available today is a free, idiot-proof Windows app that lets you download and run local AI models with a standard graphical interface.
• Nvidia PAIR: Now out of beta, this free, open-source tool can essentially turn your home into its own AI data center — allowing PCs and MacBooks to share AI compute across a local network. Basically, if your prompt needs more horsepower, it can find some from your desktop in the other room.
• Getting agent-ready: New skills are coming online so your AI agents in Hermes, OpenClaw, and Perplexity can use Nvidia’s apps (Broadcast, RTX Remix, etc.).

Also, Nvidia Broadcast is getting a big upgrade with RTX video super resolution and frame generation—both for webcam and offline file processing.

The “Nemotron” Intern

On October 15th, Nvidia is dropping the Nemotron 3.5 Bolt 74B A9B model. That’s a whole lot of words, but this is basically an improved key to agentic AI.

This is a “mixture of experts” AI agent built specifically to fit inside 24GB of memory on the RTX Spark — running constantly in the background, writing code, doing research, and monitoring apps autonomously.

Skyrim NPCs that actually talk back

That Nemotron model style has a different use too — making game NPCs have personalities. Nvidia’s built a smaller, ultra-responsive model called Nemotron 3.5 Bolt E4B (again, more alphabet spaghetti), but on October 23rd, you’ll see what this can do in SkyrimNET.

Basically, this is an AI NPC mod for Elder Scrolls V, where instead of clicking through pre-written dialogue trees, you can literally talk into your microphone, and NPCs will generate dynamic, context-aware conversations and reactions on the fly.

I can’t wait to gaslight Nazeem with this…

DLSS 5 just found another gear (and a new cheat code)

Nvidia’s DLSS 5 has split opinions in NBA 2K27 (I see you in the comments). In my mind, this 3D-guided neural rendering model works well in photorealistic games — but maybe leave it off in more artistic ones.

But on October 16th, it’s set to get a speed boost with a more efficient model that Team Green claims will push frame rates up by an additional 1.4x over what I tested. Oh, and for the millennial-coded individuals out there, Hogwarts Legacy is getting the DLSS 5 treatment in November.

Over to you: Build my benchmark hitlist

Nvidia is making massive promises about Arm-based gaming, offline AI agents that don’t charge you a cloud tax, and real-time AI video game characters. AMD claims this whole platform will falter under the weight of Windows emulation.

I have the hardware. I have the games. I have the local AI models. Let’s see what this thing can do together.

Now, while I can do plenty of testing and report on my findings soon, some of this won’t be available for a little while. So this will be an ongoing thing as models and updates become available.

More from Tom's Guide

• I didn’t believe an Nvidia RTX Spark laptop would have the same performance both on and off the charger — so I put it to the test
• Forget RTX Spark: The Surface Laptop Ultra’s magnetic USB-C charger is a game-changer that makes Apple’s MagSafe look obsolete
• AMD just flaunted its own 192GB monster chip to spoil Nvidia’s RTX Spark party — Ryzen AI Max+ Pro 495 should worry Apple, too

Nvidia has just properly unveiled RTX Spark alongside Microsoft’s flagship Surface Laptop Ultra, heralding what Jensen Huang said about reinventing the PC — massive unified memory, local AI and desktop-class power in a portable chassis.

If you ask AMD, however, Nvidia didn’t invent a new category. Instead, it validated one Team Red has been selling for over a year. Speaking directly after the RTX Spark reveal, Mike Nordquist, CVP of Client Product Marketing, didn’t hold back with me.

Between calls to celebrate industry competition and pointed critiques of Nvidia’s hardware architecture, Nordquist laid out a sharp, pragmatic case for why AMD believes its second-generation Ryzen AI Max 400 series (codename Gorgon Point Halo) leaves Nvidia’s new darling in the dust.

Editor's note: This interview has been edited for brevity and clarity.

'Nvidia is just getting started'

While Nvidia embraces the unified-memory laptop, AMD was quick to point out that it kicked off this exact architecture with Strix Halo back in early 2025.

“The exciting part is just the category development. We introduced a similar product in early 2025, so having more people come in and validate that, and you see Microsoft beefing up, I think that’s good overall. Competition is good overall,” Nordquist commented.

“We’re in our second generation that we just announced in the last month, which brings 50% more capacity [than RTX Spark]. So you can get higher capability [AI] models than maybe what you saw today.”

This cuts straight to the core of one thing AMD does have an advantage over Nvidia on — core memory. RTX Spark caps out at 128GB of unified memory, but AMD’s Gorgon Halo pushes that ceiling to 192GB. As Nordquist says himself, that gives you “bigger models up to 300 billion parameters,” which ensures a “higher quality.”

The elephant in the keynote: why did nobody say “Arm”?

Quick explainer to kick this off. AMD and Intel CPUs are built on an x86 architecture — the architecture that’s been used by Windows PCs for over 30 years. It has its down sides, but the main strength here is its the default native standard. It’s been around for ages and chips are built for it.

Nvidia’s RTX Spark and Snapdragon X Elite use an Arm architecture. Arm’s not new, but comparatively, it’s the new kid on the block which does introduce some power efficiency gains.

However, it’s main issue is emulating these apps built for x86 into Arm. There’s a performance penalty there, as it needs to translate the code as you use the app at the same time.

This is probably the sharpest jab from Nordquist. Not about raw AI performance or memory bandwidth, but the glaring omission in Nvidia and Microsoft’s keynote. Across the entire launch, the word “Arm” was practically scrubbed from the script.

“Everything just runs native [on AMD], and I still don’t know what that experience is going to be, how great the CPU is going to be,” Nordquist pondered. “We feel really good that everything just runs out of the box with out product. And I think those questions remain [with RTX Spark].”

When pressed on whether Windows on Arm is finally staging an unstoppable coup against traditional x86 computing, Nordquist dismissed the supposed architectural superiority of Arm outright.

“I don’t think there’s anything magic about going to Arm — that it’s going to be lower power or some other attribute out there. I don’t think we’ve ever believed that, and I don’t think Intel probably believes that as well,” Nordquist responded.

“I didn’t hear Arm mentioned once. It was interesting in their keynote that neither Nvidia nor Microsoft talked about it.

Thirty years of laptop DNA vs a GPU giant’s first clamshell

Nvidia might dominate AI data centers and even make the dedicated GPUs for gaming laptops. But when it comes to the messy, compromise-heavy reality of consumer laptops — thermal throttling, battery curves, sleep states and legacy software — Nordquist hammered how that building a standalone laptop platform is a different beast.

“They’re new to the laptop market in general. We’ve been doing it for 30 years-plus in this space, so I think we know how to run Windows,” Nordquist added.

“There’s no emulation; the games are just going to run natively. You’re buying into something that’s going to work, and not having to worry about whether something’s going to break. You’re just kind of covered with what we have.”

The silicon war moves to the bench

Nordquist’s retort lands with some weight because it exposes Arm’s hurdles and the challenge Nvidia has:

• The compatibility tax: Microsoft has worked hard on its Windows Prism Emulation layer for translating x86 apps to Arm, but running that translation still risks edge-case performance hits. Conversely, AMD’’s x86 architecture guarantees zero translation overhead for hefty software and native PC gaming.
• The memory ceiling: Nvidia’s 128GB unified cap is undeniably impressive, but AMD’s 192GB ceiling gives local AI 50% more breathing room for more concurrent agents.
• The category reality: Nvidia’s joined a party AMD rocked up to a year ago. The company has already spent 12 months taking consumer feedback and refining the second-generation silicon.

But ultimately, the time for talk is over. Nvidia brought serious star power, but AMD isn’t blinking. Now it’s time for real-world benchmarking to see whether Nvidia’s Arm silicon can actually match AMD’s battle-tested x86 brute force.

More from Tom's Guide

• I didn’t believe an Nvidia RTX Spark laptop would have the same performance both on and off the charger — so I put it to the test
• AMD just flaunted its own 192GB monster chip to spoil Nvidia’s RTX Spark party — Ryzen AI Max+ Pro 495 should worry Apple, too
• I just tested Windows 11 26H2 — here’s how to update now and why Microsoft finally fixed the two worst things about Windows 11

So Nvidia’s RTX Spark is officially here, and Team Green is making a big stink about its mission to “reinvent the PC” in the age of local agentic AI. But can I be honest with you, dear reader? The AI stuff is cool, but it's also fixing this one key gripe with Windows laptops, which has me believing in this reinvention.

As the headline gives away, I’m talking about getting the same performance both on and off the charger. The snail’s pace I’d get on a Windows laptop whenever it's disconnected has become a commonplace gripe of mine — so much so that I was suspicious of whether Nvidia had actually cracked it.

Well, turns out the company proved me wrong royally with some key examples over the months, showing just how efficient this is with that battery to deliver the same speeds no matter where you are.

The accident

Probably the most surprising view of this potential is in Blender. For those out of the know, this is a seriously intensive app for 3D animation and digital art — it demands a lot of your CPU and GPU in remapping pixels every time you move an image.

During IFA 2026, I saw a Blender demo that was supposed to show the app with and without DLSS activated on two identical Surface Laptop Ultras. Instead, it revealed the raw stamina of an RTX Spark system because one was left unplugged.

Speaking to Nvidia, the demo had been turned on/setup at 9 am, and I attended it at around 2:30 pm. After 5.5 hours, the 10% battery warning had only started showing.

Even on something super power-efficient like a MacBook Pro, you’d be lucky to get 4-5 hours out of it. So you can imagine how blown away I was in this situation.

On top of that, I saw Unreal Engine 5's biggest sample map running on the MSI Prestige Flip N16 without any stuttering off the charger.

No more gaming frustration

And since this is Nvidia, we have to talk about gaming. You already know the key gameplay specs here — DLSS 4.5 supported through the sheer heft of all those Blackwell GPU cores and the horsepower to run 1440p games at buttery-smooth frame rates.

Nowhere was this more obvious than with the Microsoft Surface Laptop Ultra. That magnetic USB-C port may have stolen my heart, but at the same time, I got to see Gears of War: E-Day (a game that's being translated to work with the silicon) run at exactly the same frame rate.

That standard gaming laptop frustration of watching your FPS dip is a thing of the past here. No more massive cut in performance!

What does this mean for battery life?

But let’s get real here. What is the battery life in reality? All of this sounds great, but there isn’t anything concrete beyond a claim of “all-day battery life.”

Throw in the fact that this is an Arm chip and that emulation isn’t necessarily the most power-efficient way to run apps, and this is something I’m very keen to test.

More from Tom's Guide

• NVIDIA RTX Spark laptop preorders are live: here's where to buy them
• AMD just flaunted its own 192GB monster chip to spoil Nvidia’s RTX Spark party — Ryzen AI Max+ Pro 495 should worry Apple, too
• I tried Microsoft’s new haptic Surface Mouse — and standard mice now feel completely broken

With Nvidia widely rumored to unleash RTX Spark alongside Microsoft on October 7, AMD decided it couldn’t wait. Team Red executed a preemptive strike to not just show off its shiny new (and expensive) Ryzen AI Max+ Pro 495 silicon, the company cheekily rebranded itself as “Agentic Micro Devices since 2025.”

Is this classic counter programming to plant a flag before Nvidia? Yes. But beneath the rush lies a piece of hardware with specifications so absurd they demand attention. Packing 16 desktop-class “Zen 5” cores and up to 192GB of unified memory, AMD’s built a brute-force x86 weapon designed to kill cloud AI subscription fees, take on Team Green and even make Apple a little nervous too.

The 192GB cheat code

To understand why AMD is sweating, look at what Nvidia is trying to pull off with RTX Spark. The team is aiming to make local AI agents accessible by pairing Arm-based compute cores with Blackwell graphics (a similar amount of cores that you find in the company’s desktop RTX 50-series GPUs) and unified system memory.

Apple, of course, wrote this playbook years ago with Apple Silicon in the MacBook Pro — proving that fusing a CPU and GPU around a giant pool of memory is the ultimate setup for intense tasks and heavy local models. And while Apple built the sandbox that Nvidia’s trying to take over, AMD just decided to dump a truckload of concrete into it.

If you want to run massive AI models locally on traditional laptops, the amount of memory means you simply cannot. Team Red’s Ryzen AI Max+ breaks that bottleneck through sheer volume to its digital brain:

• A 192GB shared reservoir: The CPU and GPU can share an enormous 192GB of memory running at a blistering 8.5 billion data transfers per second.
• The 160GB slider: I played with this on the Asus ProArt GoPro Edition with the AI Max+ 395. Using a feature called Variable Graphics Memory (VGM), users can carve out up to 160GB of that memory directly for graphics processing.
• The scale: This gives a single laptop more usable video memory (VRAM) than four desktop RTX 4090 cards strapped together. Not only is this huge for gaming (trust me, I tested it), but it allows for gargantuan 100+ billion-paramenter AI models to run entirely on-device and offline.

And that offline bit is critical in my mind. While I’m always going to take a look at the sheer amount of memory being gobbled up and think “oh, so that’s why RAMageddon’s a thing,” I am a believer that local AI is the future over cloud models like ChatGPT or Gemini.

Now, there are real-world examples to prove that.

Firing the cloud provider

Tech enthusiasts are tired of hearing “AI” attached to every bit of basic spreadsheet math, so AMD’s brought in Emmy-winning production studio LightSail VR to show off the economic reality of local agents.

This is the VR studio that recently caught a Sabrina Carpenter concert, and they deal with ultra-heavy 16K, 90 FPS stereoscopic 3D video for Hollywood. With client feedback pouring in across multiple projects, it was becoming overwhelming for the small team.

Option one of using cloud models like OpenAI or Anthropic to coordinate this is impossible for two reasons:

• Strict Hollywood NDAs: Confidential scripts, clips or project schedules cannot leave the four walls of their studio.
• The cloud meter: Running autonomous agents that constantly re-index documents and shift production timelines would burn thousands of dollars a month in cloud AI token costs.

So instead, LightSail VR build an in-house “production coordinator” agent on an AMD Ryzen AI Max desktop with unified memory — living directly inside the team’s personal Slack. It’s able to update schedules, rewrite spreadsheets and alert the team autonomously when an edit timeline shifts.

And the payoff has been huge, with the studio able to manage six to nine concurrent productions with the personal effort required for just two, with zero data leaving the building and zero cloud AI token invoices.

What does that mean for me?

But of course, you’re (probably) not producing 16K stereoscopic video for a Hollywood studio. Why should you care about a 192GB laptop processor? Simply put, the tech industry is currently trying to trap you in an endless web of AI subscription fees.

Right now, using a generative AI agent to organize your life or draft your emails comes with a hefty monthly premium fee. By building a laptop chip with massive unified memory, AMD is effectively stuffing that cloud server into a backpack.

That translates into two massive consumer victories:

• The death of the subscription tax: Buy the hardware once. After that, your autonomous digital intern can reorganize that messy hard drive, draft emails and analyze massive spreadsheets 24/7 without costing you a single cent in cloud AI costs.
• Absolute privacy: You no longer have to upload your tax returns, confidential work documents or personal schedules to a server in California just to get AI assistance. These gargantuan models can fit directly into the AI Max’s memory pool, so they run completely locally and offline.

It’s not just about raw horsepower; it’s a hardware escape hatch from the looming cloud subscription era.

The x86 moat against Arm

And that brings us to the core battlefield between AMD and Nvidia (and Apple to an extent).

RTX Spark leans into Arm architecture. Arm chips deliver exceptional power efficiency — as Apple has proved for years — but running traditional Windows applications on Arm requires translation layers.

And while Microsoft’s Prism emulation has improved massively (like Microsoft's engineer told me), translation layers still introduce power inefficiencies, occasional driver quirks and incompatibilities with legacy hardware.

In my experience with AMD, the team spent a significant chunk of its time exploiting that vulnerability and repeatedly mentioning “Native x86” alongside there being “no emulation layer” for Windows and Linux.

By sticking to the standard PC architecture of the past three decades, AMD avoids that trap entirely. Specialized software can run natively at full clip, older peripherals work out of the box, and when the work finishes, the integrated RDNA 3.5 graphics engine can natively fire up PC games without any chance of an emulator hiccuping mid-frame.

AMD’s scramble to share all of this two days before Nvidia takes the stage with Microsoft shows just how intense the pressure has become. Team Red is anxious that RTX Spark will set the narrative for what high-end AI laptops are supposed to look like.

But by using native x86 architecture and an audacious amount of money, AMD has ensured that when Nvidia steps into the spotlight, it won’t be walking into an empty room.

More from Tom's Guide

• I tested the Asus ROG Zephyrus G14 (2026) like I was a student — its gaming, creator and AI performance is huge, but it comes at a cost
• I just tested Windows 11 26H2 — here’s how to update now and why Microsoft finally fixed the two worst things about Windows 11
• I tested Microsoft’s fanless Snapdragon X2 Plus Surface Pro and Laptop: The MacBook Air should be terrified

Remember when the rumor mill of those AirPods Pro with cameras started swirling? We may be waiting a while longer for them, but outside of Apple’s ecosystem, Qualcomm’s just launched the Snapdragon Sound Elite Gen 2 silicon to support the devices that will compete with them.

Beyond lossless sound enhancements, ambient ANC, and enhanced voice clarity, this chip is all about multimodal AI with camera support, alongside support for agentic AI.

The big specs (to me)

So let me get straight into it, as there are some wins here. Packed into a 30% smaller footprint with 40% lower power consumption, this is bringing intelligence to premium earbuds, camera buds, audio glasses, headphones, AI wearables and more.

It starts with aptX lossless audio support, new AI enhanced ANC that adapts to your ambient surroundings, a new AI-turbocharged microphone pickup algorithm to improve voice pick-up.

This means an EQ that can change slightly around where you are, alongside local models that can let important sounds through the ANC wall (like sirens).

And for the cherry on top, there’s now integrated micro-power WiFi 6E — a world first for a headphone chip that means you could connect to cloud audio directly from the device.

The AI pivot

But let’s get into the big thing of Sound Elite Gen 2: that new local intelligence built right in. Qualcomm’s packed this with up to 2x greater AI capability than its predecessor, ready to fuel the next generation of on-camera equipped form factors (translation: all the AirPods with cameras built in for Siri vision).

Plus with that direct-to-cloud connectivity, it can support a breadth of AI services both locally and on-cloud. This will take earbuds beyond your standard voice assistant into something a little more agentic thanks to that built-in NPU and an ecosystem of apps and agents built specifically for the chip.

Do you actually want camera buds in the age of pervert glasses?

This is the big question here. My tech nerd take is of course that this is a solid silicon upgrade for upcoming devices. But we’ve come to realize a lot more about the privacy issues that come with AI smart glasses or always-on recorders.

So now that we’re seeing these kinds of devices in a whole new light — a light where we’re much more aware of how much of our lives are being used to train AI models, and the rise in covert recording — is this a case of bad timing? Would you want to buy a pair of camera buds?

More from Tom's Guide

• iPhone 18 Pro teardown reveals the ins and outs of the latest iPhone models — including one annoying repair issue
• I spent 30 minutes with the iPhone Duo — 3 things you just don’t get from the pictures
• Snap’s CEO told me he hates getting AI-drafted texts from his wife — so what’s the plan to fight brain fry with Specs?

It's September, so of course Qualcomm is back with an even more powerful mobile phone chip. However, this time, there’s two: Snapdragon 8 Elite Gen 6 and the new Snapdragon 8 Elite Extreme Gen 6!

One thing is clear, the company are keen to tell you these chips will “power the agentic AI age,” which is all fair and good (and a real advantage for this over the A20 Pro in Apple’s iPhone 18 Pro). But there are a few other places where Snapdragon-armed phones will get the advantage.

Let me explain.

Clock speeds off the charts

So it all starts with re-engineered Oryon CPU cores — the brains of the operation. And if you’re a speeds and feeds kind of person, the main thing to note here is this is the world’s first 5GHz mobile CPU. Those Prime Cores on 8 Elite Gen 6 are sure to tear through even the most intensive of tasks.

Now, this is of course before any real benchmarking has happened. Apple has always been up to some serious witchcraft and wizardry when it comes to this wicked single core speeds. But that symphony of Prime Cores will surely introduce some bonkers multicore speeds and give Snapdragon the edge over A20 Pro.

Bringing DLSS to smartphones

There’s a 12% power improvement and new AI-dedicated GPU cores on-board here. Beyond all your usual accelerated local AI numbers, one of the big things that comes with this is Adreno Neural Fusion — AI super resolution and frame generation for games.

Time will tell on how many games will support this tech looking ahead. But with a 40% improvement in power and performance in key titles like Naraka Bladepoint and Honkai Starrail, as well as integration into game engines for future ease of support, this could be significant for bringing more AAA titles.

My Sensing Hub is tingling

Built onto the chip is what Qualcomm’s calling the Sensing Hub. For local AI to be at its most effective, it needs a chance to be ready to listen at all times, support suitably sized models to understand all your personal context, and zippier speeds with greater efficiency.

This is the thing that ties that agentic AI loop through the CPU, GPU and NPU, as ambient voice is captured, and with low latency, responses are received entirely on-device. A20 Pro is absolutely capable of this, but 8 Elite Gen 6 is taking it a step further with a dedicated neural chip alongside the bigger NPU.

Camera ceiling lifted

And last but certainly no means least, we get into the camera resolution. Any slab with an 8 Elite Extreme Gen 6 inside it has an advanced Spectra Image Signal Processing unit (ISP) that is capable of handling 8K 60 FPS video recording — or 4K 240 FPS slow motion.

Now, bear in mind that I said “resolution” and not “quality.” The iPhones will always have a spectacular camera system, but if extra sharpness matters to your captures, 8 Elite Extreme phones will be capable of it.

The similarities

But the more things change, the more they stay the same. For every difference I can point to here, there are other ways these two SOCs are pretty similar, such as AI accelerated GPU cores, a combo of performance and efficiency, and a secure enclave built directly onto the chip.

There are some differences across these tiny dies with transistors thinner than the average human hair, but what they unlock seem to be similar enough: boosting NPU and memory throughput for a local AI future, turbocharging the GPU and further squeezing that power efficiency,

Look forward to tests from me soon enough!

More from Tom's Guide

• iPhone 18 Pro teardown reveals the ins and outs of the latest iPhone models — including one annoying repair issue
• I spent 30 minutes with the iPhone Duo — 3 things you just don’t get from the pictures
• Snap’s CEO told me he hates getting AI-drafted texts from his wife — so what’s the plan to fight brain fry with Specs?

Apple has just debuted the M6 chip in its brand new Mac mini, and it’s the world’s first commercially available 2-nanometer (2nm) chip. That means every transistor is roughly 40,000 times thinner than the average strand of hair, which means the Cupertino crew can pack many, many more transistors onto the chip die.

That means higher clock speeds and massive energy efficiency gains without the thermal penalties that normally come with it. But with that additional room for activities, Apple’s done something rather interesting here. I’ve combed through every tiny detail the company’s revealed about M6 to take a deep dive into the silicon.

And spoiler alert: up until now, I’ve been telling my friends and family with M1 systems to hold out over the past few years. But this is the year I’ll finally tell them to upgrade.

M6 chip specs comparison

The rise of the super core

In previous generations, Apple’s used a classic two-tier layout — performance cores for handling the heavy lifting and efficiency cores to tackle the day-to-day that doesn’t require breaking a sweat.

M6 does things differently by bringing on a 12-core CPU complex (up from 10 cores on M5) that is split across three tiers:

• 2 Super Cores: Engineered for the moments of maximum burst speed, Apple claims to have built the world’s fastest CPU core here.
• 4 performance cores: Same as before, these are there to tackle heavy, sustained multithreaded tasks like data analysis or music production.
• 6 efficiency cores: These are on-board for handling all the background macOS processes and general day-to-day multitasking with minimal wattage.

And the numbers (claimed by Apple) speak for themselves: up to 1.2x faster multithreaded performance over M5 and a staggering 2.4x leap over the base M1.

Doubling down on local AI

Remember when everybody bought an M4 Mac mini to run their OpenClaw? Apple remembers, and instead of a single NPU block, M6 is debuting a dual 16-core Neural Engine that can deliver up to 2x peak AI compute over previous generations.

Alongside this, the upgraded 12-core GPU is packed with dedicated Neural Accelerators inside each core for a 30% boost in peak AI GPU compute over M5 (and 8x over M1). Together, these ensure rapid prompt processing, and accelerated local LLMs and agentic tasks.

And another word on that graphics muscle, there’s 2nd generation Dynamic Caching, hardware ray tracing and a 50% increase in geometry rates. That’s a word soup, but the end result is going to be much faster graphics rendering — good news for gamers.

Feeding the 2nm beast

Any new chip is only as good as the memory feeding it, and for M6, Apple is raising the ceiling up to 32GB of unified memory in the Mac mini. That’s going to ensure bigger AI models or more sustained multitasking is possible.

But moreover, the bandwidth of that memory (how fast the chip can dip into there and take something) has reached 170 GB/s — a 10% uplift over M5 and a 2.5x surge compared to the original M1’s 68GB/s ceiling.

Put simply, it’s like giving your Mac mini a much bigger workbench to craft things without anything spilling onto the floor, as well as a massive high-speed multi-lane conveyor belt to rush raw materials to that bench.

The consumer verdict: who actually needs M6?

It’s a good question, and now more than ever, price matters. At the end of the day, the new Mac mini starts at $899 — a $100 uplift from the previous model. So who should jump on this train?

• Skip if you have M4 or M5: A 20% multithreaded bump and tighter AI execution are great, but not worth abandoning a machine purchased in the last 12-18 months.
• M3 falls into a nuanced middle ground: M6 may bring big gains, but the M3 still holds up by being the first to introduce a lot of what makes this new chip strong like the zippy neural engine or blazing single-core performance. Day-to-day tasks won’t feel night-and-day different, but if you demand sustained multithreaded horsepower (or you want to run local AI via that dual Neural Engine), the M6 provides the headroom you’ve been missing.
• A strong upgrade from M2: The jump from M2 to M6 spans four years of architectural pivots, a huge gap in hardware features (like hardware ray tracing a GPU neural accelerators) and those super cores. If you’re coding, using creative software or pumping local AI, the leap in single-core performance with those new Super Cores makes this feel like a new machine.
• The ultimate M1 & Intel upgrade: With 2.4x CPU performance, 8x AI GPU acceleration and 2.5x memory bandwidth, M6 represents the ideal leap for users still clinging onto original 2020 M1 Macs or legacy setups.

For anyone still nursing an M1 machine through its sixth year, the wait is officially over — the 2nm era is where you finally get off the fence.

More from Tom's Guide

• Is the MacBook Neo a good back-to-school laptop or a trap?
• Tom’s Guide Computing Awards 2026: 21 game-changing laptops, desktops and peripherals
• I just got invited to the Googlebook September 15 event in NYC — this is what I expect to see

Nvidia's ARM-based gaming laptops powered by the RTX Spark is around the corner and appear to be ready to launch thanks to new Geekbench listings.

The new ARM-based CPUs were first revealed at Computex in May by CEO Jensen Huang. "Our focus is to reinvent the PC. Windows is of course improving, but the basic architecture of a PC has largely been the same now for about 40 years, and we want to reinvent it," Huang told Tom's Guide.

How that looks is starting with an Arm-based design over the conventional x86 architecture, but Nvidia does have experience in that realm. The company already makes Arm gaming chips that power the Nintendo Switch 2.

The RTX Spark platform is starting out on a range of Windows gaming laptops and mini PCs with a focus on gaming and agentic AIs. It should have fairly low power consumption.

Geekbench scores revealed

Two different Geekbench listings reveal results for the RTX Spark N1X processor in 20-core and 18-core configurations.

If we focus on the 20-core configuration, it returned surprising scores with a single-core of 2,593 and a multi-core score of 23,424. The multicore is fairly impressive, comparing favorably with Intel's i9-14900KS, which managed a score of 23,785 in the same test.

View the full results here, though I will note that the final production scores might differ.

If you're unfamiliar, Geekbench measures how well the processor handles everyday tasks by running a bunch of simulated workloads from processing numbers to crunching numbers.

Single-core is all about how fast the CPU handles one task at a time, like editing a photo. Multi-core kicks in for multitasking like running several apps or rendering by using all the chip's "cores."

Here's how those numbers compare to Apple's M5, the newer Qualcomm Snapdragon X2, and Intel's Panther Lake chips.

Surprisingly, the RTX Spark performs well against the M5 and Intel X9 in multicore. Moving up to the M5 Pro and Max gets you higher scores. The same goes for the higher-end Qualcomm Snapdragon X2 Extreme, which has recorded over 25,000.

Still, the RTX seems like it will compare favorably with most every chip on the market.

Where it seems to fail is in the single-core, where scores are significantly lower. It seems it might feel sluggish in those single tasks, but running multiple apps is where the RTX Spark might shine.

This is a pretty good indication of how the RTX Spark platform will handle CPU-heavy tasks. It doesn't tell us the whole picture, especially when it comes to gaming, but we do know the GPU will support DLSS 4.5.

RTX Spark laptops are supposed to arrive later this fall with Windows 11 out of the box. It should be compatible with most Windows games, and Nvidia is reportedly ensuring the chip will work with common anti-cheat and DRM software.

Nvidia has already confirmed that it will launch with laptops from Asus, MSI, Microsoft and more.

More from Tom's Guide

• Nvidia RTX Spark is here, and it's going to 'reinvent the PC' — here's everything you need to know about this all-new laptop chip
• Intel Arc G3 is here to shake up handheld gaming, and AMD should be terrified — Acer Predator Atlas 8, a new MSI Claw and more launching soon
• I tested the Intel Core Ultra 7 270K Plus and 5 250K Plus against AMD’s best gaming CPUs — this should be a mismatch, but the results shocked me

The biggest news at Computex is official, as Nvidia’s CEO Jensen Huang just announced RTX Spark Super Chip family — the company's first-ever all-in-one laptop silicon that takes on Intel, AMD, and Qualcomm in a big way with a GPU equivalent to an RTX 5070.

Set to release this fall, this is not your ordinary ARM chip. The RTX Spark is packed with CUDA cores, which is sure to make that integrated GPU one helluva gaming monster. Plus, Spark is set to change the way you interact with Windows PCs forever with an AI agentic twist. I will be testing the chip extensively during my time out here at Taipei, and talking to a couple of the engineers behind it.

But for now, I’ve taken a deep dive into the specs tables and found what could be a breakthrough moment similar to when Apple brought the M1 to the world back in 2020.

Nvidia RTX Spark Super Chip specs

Nvidia RTX Super Chip devices

Nvidia is expecting a huge lineup of systems to launch with RTX spark — with plans to release over 30 laptop models and well over 10 desktop variations, targeting a range of price points.

But as of now, there are a whole lot of devices launching this fall with the RTX Spark Super Chip inside. On the laptop side, we have:

• Asus ProArt P14 and P16
• Dell SPX 16
• HP OmniBook
• Lenovo Yoga Pro 9
• Microsoft Surface Ultra
• MSI Prestige N16

These devices will be as thin as 14 millimeters, and as light as three pounds — a serious upgrade over the chunky Nvidia gaming laptops you're used to.

But they're not stopping at laptops. In an announcement that probably woke up Anthony Spadafora (our mini PCs king) from his nap, RTX Spark will be coming to small form-factor desktops that are "small enough to disappear on a desk."

These will be made by Dell, HP, Lenovo, Asus, MSI, Acer, and Gigabyte.

Explain it with pizza

This is partly inspired by the tasty ‘za I found while typing this, but also shoutout to Nvidia for specifically pointing out that I just “love to explain stuff with pizza.” You asked for it, and now you’re getting it!

In the laptop space, you’ve got various ways of serving pizza.

• The standard meal (an x86 SoC): This is like ordering a plain cheese pizza. It works perfectly fine for everyday eating, but if you want heavy flavor (think intense gaming or AI generation), it lacks all the necessary toppings. Don’t get me wrong, Intel and AMD have been adding some new garnish, but they’re still a little hampered by architectural limitations.
• The combo meal (x86 CPU + dedicated GPU): For that premium flavor, you have to order the standard pizza plus a massive side of premium toppings. Passing these ingredients back and forth across the table does the job, but it takes up way more space (think thicker gaming laptops) and requires a whole lot of power.
• The efficiency slice (Snapdragon X Elite): Qualcomm is already dealing in making efficient single-crust pizza using a similar recipe (Arm architecture). It’s great for saving power, but Nvidia's world-famous secret sauces (CUDA and RTX) can’t be matched.

So in this backdrop, I can only describe the RTX Spark Super chip as a “super pizza.” Instead of ordering a meal with lots of separate sides, this is the all-in-one slice baked onto a single, highly efficient crust. Half of the pizza is that custom CPU, and the other half is loaded with incredibly spicy, high-end toppings in that Blackwell GPU.

But the magic is in the cheese (the memory). Instead of dividing the cheese between the crust and toppings, Spark uses up to 128GB of “unified cheese” that the entire pizza shares instantly at blazing speeds (five times faster than standard PCIe Gen 5). And because it’s baked so efficiently, you can take this super pizza anywhere without it getting cold (all-day battery life).

…am I stretching this analogy? Absolutely. Do I regret it? No (hi Nvidia if you’re reading this).

Going agentic

With this increase in RTX Spark's brute force, Nvidia is working with Microsoft to shift personal computing on Windows from a passive tool to a proactive agent.

This is a one petaflop AI superchip... For context, some quick math I did while sitting in the keynote, the AMD Ryzen AI Max+ 395 can do around 60 Teraflops. This thing is built differently. To go back to the pizza analogy, not only can it make a mean slice, but it can probably feed it to you, too.

It can work across apps on your device — stringing together different tools like any user would to get the job done. For example, an artist can provide a sketch and a mood board, then use Photoshop not only to generate the art locally but also to turn it into a 3D image and animate it into a video. And this is all done without ever needing cloud compute.

Or if you’re a gamer, you can use G-assist to tweak critical settings like monitor optimization, or (if you’re streaming) tell it to operate your lights, mute the mic, and turn on your be right back scene in OBS.

Essentially, it can take complete control of your system — even the mouse and keyboard to navigate as a human user would. Keeping this all secure is Nvidia OpenShell, which brings enhanced security to ensure your model use works locally and privately. It can even mask personal information before ever using a cloud-based AI model to do anything.

For the players

But of course, there are big gaming capabilities here too. This is Nvidia, after all. Team Green says this falls into the same performance class as an RTX 5070 laptop GPU but uses significantly less power to reach that level.

That means you could be playing the likes of Cyberpunk, Doom, or Indiana Jones at 100 FPS at 1440p resolution. And given its ARM, the RTX Super Chip is capable of this level of performance both when plugged and unplugged.

Of course, battery life will vary based on how much you throw at it, but when I asked Nvidia, a tweak of maximum frame rate and graphics settings can optimize this nicely for long gameplay sessions on flights.

Now, what about the emulation part of it? General apps are included in this, too, but gaming has been the big test for Windows on Arm, and that Prism emulation layer used to translate apps written for other CPUs to this new architecture. Well, Nvidia’s thought of that too.

The company is working with a large list of developers to either port games natively to Windows on Arm or ensure they run well through the emulator. And with that hefty GPU, anything that’s particularly taxing on that will have almost no negative effect on performance.

Plus, Nvidia has collaborated with all first- and third-party anti-cheat software providers to ensure competitive multiplayer games launch without a hitch.

Nvidia RTX Spark vs Apple M silicon vs Snapdragon X2 Elite

So, since I’m calling this Nvidia’s M1 moment, how does it stack up with Apple laptop silicon? While I’m at it, I'm gonna throw Snapdragon into this comparison too, as they’re all similar-ish in Arm architecture.

But the crucial differences show that it’s not about the tools you have, but rather how you use them. You see, Apple silicon isn’t just about the brute force of the hardware; it’s also about its tight integration with software that extracts that performance. In gaming, for example, the MetalFX layer has enabled MacBook Pros to do some pretty great things, like run Cyberpunk 2077.

And while Snapdragon X2 Elite is level pegging in certain benchmarks, there are still some moments where it can feel like a bit of a blunt instrument against Windows 11 — especially in that GPU department.

But with RTX Spark, this is Nvidia we’re talking about. That integrated GPU supports DLSS 4.5, so you can bet your bottom dollar that the out-of-the-box gaming performance is going to be clearly superior to Apple and Snapdragon.

On top of that, Team Green and Windows are rewriting the rules on local agentic AI, so you’re getting a heap of AI coworkers too, with those CUDA cores for the raw brunt.

Nvidia RTX Spark vs Intel Core Ultra Series 3 vs AMD Ryzen AI 400 series

Why did I put these separate from the face-offs with Snapdragon or M5? Well, it comes down to the different ways these chips compute stuff. Nvidia, Apple and Qualcomm all have Arm architectures, which do things very differently to the x86 chips from Intel and AMD.

That being said, the 30+-year-old x86 architecture does bring some material benefits over ARM — namely, zero app compatibility issues whatsoever. It’s a challenge that Microsoft and Nvidia are overcoming with a separate Arm-specific version of Windows 11, alongside working closely with developers. That could narrow the library you can play!

However, for what it may lack in this area, it absolutely makes up for in a sweeter blend of performance and power efficiency. While x86 is all about multitasking its way to doing things you ask it to (not the best on that battery), ARM tackles one task at a time very quickly. That’s easier on the power while bringing high speeds to the table, too.

Outlook

It’s been a long road since the first time we heard rumblings about this chip. In fact, I even put this in my Computex 2025 predictions! But now it’s here, I think it’s fair to say that RTX Spark is far beyond any expectations I had.

Not only is it a beastly integrated GPU for monstrous performance and impressive power efficiency (especially in gaming), but the software tie-ins fundamentally change how a Windows laptop works with you. This is going to be an interesting few months as we count down to the fall launch.

Now the wait is on for an Nvidia RTX Spark gaming handheld…that would slap so hard!

More from Tom's Guide

• I just tested the MSI Claw 8 EX AI+ — Intel Arc G3 is a breakthrough for handheld gaming, but at what cost?
• Intel Arc G3 is here to shake up handheld gaming, and AMD should be terrified — Acer Predator Atlas 8, a new MSI Claw and more launching soon
• I built a touchscreen MacBook Pro using this snap-on accessory, and I'm baffled why Apple hasn't made its own for years

In all my time testing Panther Lake, it’s been clear that Intel Core Ultra Series 3 chips are huge for gaming — packing a beasty internal GPU with all that XeSS 3 AI trickery to boost frame rates to unseen levels in anything outside of a dedicated Nvidia RTX 50-series gaming laptop.

Team Blue knew they were onto something, and confirmed a different version of this chip with that same GPU will be coming to gaming handhelds. And now at Computex 2026, it’s finally happening. Meet the Intel Arc G3 and Arc G3 Extreme processors — you can expect full testing and a deep dive from me soon, but for now, here’s everything you need to know.

Intel Arc G3: Specs comparison

The deep dive

A lot more details (and hands-on time) will be shared at Computex, and you'll see these handhelds start to roll out in June, with broader availability "throughout the year."

Translation: expect to see Arc G3 and Arc G3 Extreme handhelds launch sometime in August.

These chips take that breakthrough integrated GPU from the Core Ultra X series chips — the 10-core Arc B370 for the G3, and the 12-core Arc B390 for the G3 Extreme — and stuff them into a handheld with a lower wattage.

The result? I predict shockingly good gameplay on the go with power efficiency at its heart (thanks to that improved power management and the new efficiency cores being able to take on more of the load).

But the advantages over AMD keep coming. Intel is able to precompile in-game shaders using cloud servers (no more waiting ages for your cloud shaders).

Oh, and XeSS 3 is the big thing here, with AI-based resolution scaling, multi-frame generation up to 4X, and low-latency tech to ensure those AI frames don't bog down your gameplay responsiveness.

What handhelds will pack Arc G3?

At the moment, we know of three.

Acer Predator Atlas 8

As the name suggests, it’s an 8-inch handheld that sports this new silicon to support ray tracing alongside all that XeSS 3 AI-powered goodness. To keep it cool, there’s the Predator AeroBlace thermal management with the first metal fan in a handheld for up to 10% better airflow.

“Handheld gamers want PC-class performance without being tied to a desktop or charger,” Jim Johnson, SVP and General Manager of Client Computing at Intel, commented.

“With Intel Arc G-Series processors and Acer’s Predator Atlas 8, we deliver smooth, high-fidelity gameplay with exceptional battery life in a form factor you can take anywhere.”

Speaking of not being tied to a charger, the Atlas 8 sports an 80Wh battery (for the G3 Extreme — the G3 gets a 60Wh cell), and games are sure to look pretty on that 1920 x 1200 panel with 120Hz refresh rate and a variable refresh rate.

As for the ergonomics, you've got full-size analog sticks, and (what looks like) well-placed buttons directly alongside them, as well as Hall Effect triggers for all your pressure variations in racing games (plus a micro-switch mode to make them instant clicks for shooters).

The others

However, when Intel made the announcement on stage at CES 2026, other brands were connected to it, and Team Blue has confirmed the MSI Claw 8 EX AI+ and OneXPlayer will be launching too.

So all eyes are on Computex as more are inevitably revealed in Taipei! I’ll update this piece when we know more.

Intel Arc G3 vs AMD Ryzen Z2 Extreme

On paper, if you were to look at the specs table, the numbers on that GPU are bigger on AMD’s silicon. But as I found out testing Panther Lake, it’s not about what you have, it’s about how you use it.

XeSS 3 has already blown my mind in a laptop, to the point that I begged Intel way back in October to bring this tech to the handheld space. The secret sauce is that resolution scaling and 4x multi-frame generation that extracts so much more performance-per-watt out of these machines. I tested it (in a laptop) against Nvidia’s DLSS 4, and my mind was blown.

At 1080p, there’s definitely a gap in frame rates, but nowhere near the kind of gap you’d expect in integrated vs dedicated graphics. To get something akin to this on something as small as a handheld is going to be a truly generational shift.

Throw in the fact that you’re only getting 44ms of latency on Intel vs Nvidia in-game, and you start to see just how earth-shaking this could be.

Now, what about AMD? Well, the Ryzen Z2 Extreme was a minor uplift over the Ryzen Z1 Extreme — it uses an RDNA 3.5 architecture, and at the moment, its own FSR resolution scaling and frame generation doesn’t hold a candle to XeSS.

That being said, Team Red did recently say that its latest AI-fueled FSR upscaling 4.1 will be coming to RDNA 3 GPUs. In July, this will be limited to just the desktop-class RX 9700 card, but given the cores are the same, this is AMD’s big chance to keep up.

Bottom line

You can expect more hands-on testing from me in the very near future as I head to Computex. But for now, it's fair to say that if you’ve been waiting for an actual next-generation handheld (maybe you’ve been holding onto that Steam Deck for a few years now), it’s happening now.

Intel Arc G3 and G3 Extreme are really bringing the heat in terms of a monstrous integrated GPU and XeSS 3 to deliver what will be one of the best handheld gaming experiences you can get.

One thing we don’t know is price, and given the massive price increase of Steam Deck recently, that makes me nervous. But time will tell on this around whether the price-to-performance is still there.

More from Tom's Guide

• Computex 2026 predictions: I'm hoping for Nvidia N1X, Intel Arc G3 handhelds, a new Asus ROG Ally and the industry's fightback against RAMageddon
• I can finally turn my MacBook Pro into a gaming laptop (sort of) — GameHub just fixed my Mac’s biggest weakness in minutes
• I’ve driven over 10,000 miles in Forza Horizon 6, and can confirm it is a dream game for car lovers

For three years, if you wanted a great desktop CPU for gaming, you bought AMD, period. Specifically, anything with “X3D” on the end of the name. And it’s been that way because Intel’s previous shot at desktop chips (the original Core Ultra Series 2 or “Arrow Lake”) dropped the ball.

There were performance and power issues, an abnormally high latency between the chip and memory, and all of this led to gaming performance that fell behind even Intel’s older CPUs. One thing was clear: Team Blue had to go back to the drawing board.

And they’re back with what seems to be a refresh on paper — the Intel Core Ultra 5 250K Plus ($199) and Core Ultra 7 270K Plus ($299). Strong pricing, but I was skeptical about what “plus” actually meant.

So I did what any normal person would do: I put them in an insane mismatch scenario against the AMD Ryzen 7 7800X3D and Ryzen 7 9800X3D. With more L3 Cache memory (AMD brands it as 3D V-Cache), you can always rely on these for strong, smooth game performance.

But Intel’s found a cheat code. Starting with the fundamentals of speeding up the Die-to-Die frequency (the travel time for data inside the chip), and optimizing it with an all-in-one Intel Platform Performance Package (IPPP) that brings so much more potential out of it, these chips come too close for Team Red’s comfort.

Don’t get me wrong, if gaming is your priority, AMD still takes the win. But only by a whisker, which is a concern because Intel’s chips are so much cheaper — the Core Ultra 5 being $250 less than the 7800X3D, and the Ultra 7 $180 less than the 9800X3D. And that’s before even getting into the data that shows the Core Ultras being fantastic all-rounders. But I digress, let’s get into it.

How Tom’s Guide reviews CPUs

We already test pretty much every CPU that’s announced in many different ways — either packed into one of the best laptops or being the mega brain inside the best gaming PCs.

But this is the first proper CPU review Tom’s Guide has done in its many years of existence. So, as Managing Editor of Computing, I’m gonna be upfront and set the ground rules on what to expect from CPU reviews from us. It all starts with answering the three key questions anybody has before buying one:

• How well does it play my games?
• What work can I get done with it?
• Is the price (short-term and long-term) fair for what I get?

That means taking a whole lot of testing, deciphering the advanced architectural and software processes happening here, and boiling them down to what they mean for you: the user. It won’t be complex, and that’s by design because I want to answer these three questions for everyone. Whether you’re an experienced PC builder, someone getting into making your first PC, or you’re buying a pre-built and making sense of the spec list.

If you’re looking for a super-dense level of detail, this is where I point you: my friends (and old family — used to work on their team) at Tom’s Hardware! This review is for the rest of us.

Intel Core Ultra 5 250K Plus & 7 270K Plus: Cheat Sheet

• What are these? These are a new line of Intel desktop processors — a refresh from the last gen that addresses a lot of the previous issues.
• Who are they for? Enthusiasts and creators who want a great all-around balance of productivity and gaming.
• What does it cost? Final retail pricing may vary (thanks, RAMageddon), but MSRP at the moment is $199 for the Ultra 5 and $299 for the Ultra 7.
• What do we like? These chips take a massive multi-threaded lead in productivity over AMD’s X3D chips in productivity and content creation, gaming performance has been upgraded nicely, and everything runs surprisingly cool.
• What don’t we like? Intel's use of a complex software stack like this performance package means you’re not getting peak power out of the box (a couple of extra steps to get it), and 1% frame-rate lows don’t match AMD’s hardware-cache-heavy design.

Intel Core Ultra 5 250K Plus & 7 270K Plus: Specs

Intel Core Ultra 5 250K Plus & 7 270K Plus: The ups

You could see what Intel was trying to go for with the O.G. Arrow Lake — a balance between work and play, which it didn’t deliver on. This is Team Blue’s comeback moment, and they’ve taken full advantage.

I’ve tested all four in their best possible light — enabling XMP for Intel and Resizable BAR for AMD. More on those in an upcoming piece because you could be leaving a lot of potential performance on the table if you don’t turn them on! This will cause some differences on a motherboard-by-motherboard basis (I used the Asus ROG Maximum Z890 Apex), and RAM speeds will give Intel a slight leg up. But as somebody who drops a lot of cash on your PC, these will be the settings you’ll go to to make the most of what you buy, so it’s only fair to do the same here.

Aiming for the king

So Intel calls this its “fastest gaming processor.” Let’s put that to the test, shall we?

What is fascinating here is that these are actually beginning to trade blows with X3D chips, and you can see that when put into real-world gaming performance, too.

Now you may be looking at the frame-rate gap AMD has over Intel, at Intel’s lead in a couple of 3DMark synthetic benchmarks, and thinking “WTF!?” Well, that’s where we’ve got to go a bit geeky:

• Monolithic vs tiled: AMD’s X3D chips use a single monolithic chip with 8 cores. This means all the cores can immediately communicate with the memory and the GPU with very low latency. Intel uses a “tiled” architecture — basically different cores of the chip do different things, and while that Die-to-Die (D2D) frequency is 900MHz higher, that’s still a longer time for cores to talk to each other, which can lead to lower frame rates (even when the synthetic speeds may be higher).
• The Cache gap: AMD’s chips pack 96MB of L3 cache (known as the 3D V-Cache), which acts as a massive waiting room for game data like physics and NPC logic. The Core Ultra 7 270K only has 36MB, so even though your chip cores are technically faster, it still needs to sort out an overflow waiting room on the system RAM, which is going to be slower.

There are situations where this can be felt most (more on them later), but the big thing here is that Intel is back on the right track in terms of gaming performance, with nice gains over previous-gen Arrow Lake chips and closing the gap on AMD.

Beasty productivity performance

Same as the previous chips, these are productivity monsters. Whether acting alone or as a conductor to the GPU, the performance and efficiency core combo makes better sense of both day-to-day tasks and intense workloads to deliver zippier speeds.

So whether it’s pushing local AI to its limits (currently training a version of myself…it’s getting weird), complex 4K renders or just simple multi-tab web browsing, these won’t let you down.

And it’s also being done while they keep their cool much better, too. At peak stress of a multi-thread render, you can see that the Core Ultras sit far prettier at around 154-degrees fahrenheit, while the thermal density of putting 3D V-cache atop AMD CPUs like a blanket of sorts does spice things up a little.

Solid bang for your buck

But going band-for-band misses the bigger point, which is the lower price here. Yes, AMD is still ahead in terms of gaming, but with that gap narrowed and the price of Intel’s options being kept quite lower, the performance-per-dollar swings to Team Blue’s favor.

And Intel’s able to do this with some key technologies that allow the chip to work smarter, not harder. Before the IPPP, you can imagine the CPU as a chaotic kitchen — getting tons of orders in and not necessarily knowing what order to tackle them in. The salad gets soggy under a heat lamp while the steak is raw, and Gordon Ramsay is kicking off somewhere about it.

But the IPPP pipeline acts as an executive chef standing at the pass and has a set number of steps to go through:

• The Application Optimization layer (APO): The kitchen has been trained on a small number of dishes, so when the chef sees a rush of orders for the “Wukong Special,” they know exactly what to do. So far, APO supports 21 game titles — shame that it’s so small, but only up from here.
• The Binary Optimizer Tool (IBOT): Instead of letting the cooks (the CPU cores) work randomly on different dishes, the Executive Chef intercepts the tickets and tells each cook to work on all of one thing at once. Machine code gets translated and streamlined in real-time to unlock better performance.
• Die-to-Die frequency boost (D2D): By increasing this by +900MHz, the executive chef just installed a faster conveyor belt in the kitchen, so the food moves quickly to the dining room (the GPU and RAM) without any bottleneck.

This is all opt-in, and I highly suggest you do to extract the most bang for your buck.

Intel Core Ultra 5 250K Plus & 7 270K Plus: The downs

But while Intel has royalty in its sights, there is one crucial place where it does fall down to AMD’s X3D silicon supremacy.

The 1% Low reality check

Testing game performance is much more than just telling you the average frames per second — you’ve got to look at the 1% Lows.

The intensity of graphics can turn on a dime from just looking at a plate on a table to a chaotic multi-person action scene, and while you can see the best FPS on average, measuring the average of the lowest 1% of frame rates shows you whether you’re in for a bit of stuttering in certain games.

For all the software trickery Intel’s done great work with here, it can’t overcome a pure hardware advantage. AMD CPUs have up to 3X the amount of memory cache on the chip, which forms a large hard-wired safety net for smoother gaming by putting essential instructions directly on the chip.

Meanwhile, the Core Ultra 5 and Core Ultra 7 suffer from a roughly 10-15% drop in 1% lows in comparison to AMD, leading to some small stutters that can be a problem in competitive titles.

Intel does have a couple of key things going on here to try to nullify this advantage — a fix to the latency with a 900MHz speed boost of Die-to-Die frequency (the speed at which data is moved around) and that software counter-attack with Intel’s Binary Optimization Tool (streamlining machine code in real-time to reclaim performance). This makes them much better than the Arrow Lake of old.

But Intel’s IBOT has a tiny bit of overhead, and if game logic gets chaotic, the instant response of a boatload of L3 cache is always going to win out. It’s not something that will overtly impact your gaming experience if you’re more of a single-player/cinematic player.

Intel Core Ultra 5 250K Plus & 7 270K Plus: Verdict

Intel aimed for its all-rounder era before and failed. Now, they’ve only gone and done it! Are they going to take the X3D’s crown for gaming? Not with those reduced 1% lows if you’re feeling competitive online. But for the prices that Team Blue’s charging, does it really matter? Not particularly.

Whether it’s in lower-priced pre-builts compared to AMD systems or in planning out your own DIY setup, this is a better balance of capabilities that makes these great CPUs for heavy work and cinematic play.

Windows PC makers are struggling through the ongoing RAM crisis, and a new report indicates it's not going to get better, thanks to an integral PC component getting more expensive.

The Korean outlet, ETNews is reporting that Intel will raise the prices of its CPUs by 10%, largely because of skyrocketing RAM prices due to AI greed. The publication says Intel informed its major clients of the price hike today (March 19) with increases expected to hit "most" of the company's CPU lineup.

It's suggested that the move is being made to secure profitability as the supply of consumer devices (ie gaming PCs, laptops, etc) shrinks due to the ever-growing demand of AI data centers.

Beyond RAM prices, Intel is facing stiffer competition from AMD and Intel. While Intel still dominates the CPU market, holding 60% of market share overall, AMD is catching up. According to charts from CPUbenchmark.net, Intel makes up 73% of Windows laptop share, but was surpassed by AMD last year in desktops at 51%.

This year, Qualcomm announced a new Snapdragon X2 Plus CPU that is supposed to be an "epic" upgrade. At CES 2026, AMD claimed its Ryzen AI CPUs are faster and better than Intel's Panther Lake for both gaming and AI.

Intel may be feeling the heat.

'Difficult to survive'

"There are concerns that if Intel CPU prices also rise, operating profits will shrink significantly, making it difficult to survive," an industry insider told ETNews.

Still, any price hike during RAMaggedon is going to adversely affect PC manufacturers who are struggling in the face of new competition from the $599 MacBook Neo and surging RAM prices.

Outside of the Neo, we may be seeing the end of the era of cheap PCs, with no relief insight until 2028, at the earliest. PC makers are trying different things to survive from seeking out smaller memory producers to focusing on premium PCs over budget offerings.

For now, the RAM crisis just keeps claiming victims. Even major companies like Samsung have put its mobile division into "emergency management" mode despite record pre-orders for the Galaxy S26 Ultra.

More from Tom's Guide

• Intel’s Core Ultra 200S desktop CPUs are 2x faster and $100 cheaper — a second chance at Arrow Lake's first bad impression
• Best graphics cards: I've tested every GPU to find the best bang for your buck
• I tested Intel XeSS 3 vs Nvidia DLSS 4: my laptop tests made me rethink the future of integrated graphics

As inevitable as day turns to night, the latest Apple silicon tests get leaked on Geekbench. This time around, it’s the M5 Max that has been exposed and as you probably expected, it’s an absolute monster.

But it’s easy to just see the M5 Pro and M5 Max MacBook Pros as simply “big number goes up” — I mean that’s exactly what my mates thought when I was trying to explain the new architectural changes while sharing a pizza.

“So, what’s the big deal then?” Ben asked (with an inappropriate amount of tomato sauce around his mouth). Then, I looked at the slice I was holding, and it hit me.

Apple calls it “Fusion Architecture,” and this is a pivotal moment where the company has stopped trying to make a bigger pizza, but build a buffet instead. This isn’t just a small change, it’s a generational leap that changes the way your computer thinks from having a single brain to a hive mind that can scale as big as the buffet allows.

Sounds weird, I know. But let me explain.

Why the monolith had to die

Intel names it “chips on a system” for its Core Ultra Series 3 chips — whatever it’s named, this is the direction everyone is heading in. And in trying to explain this to my friends, I went to the one dish I’m always thinking about.

Imagine you’re making a pizza. As more and more people love the pepperoni special you make, you have to go bigger. But you can only make a tasty ‘za so big before it won’t fit in the oven, or the middle stays raw while the crust burns.

Traditionally, Apple Silicon has always been a system on a chip (SoC): everything (the CPU, GPU and RAM) on one giant pizza. Great for efficiency, but limited in being able to scale up.

Chipmakers cannot print a single chip larger than a certain size without it becoming buggy and incredibly expensive. There’s a scaling issue (roughly 858mm-squared based on maximum exposure area of the lithography machines that make chips) and a quality yield issue:

• You could keep making the dough bigger, but your pizza oven is a fixed size.
• If you have a 5-foot monster pizza and a single hair falls on it, you have to throw the entire thing away.

So instead of trying to make one impossible pizza, Apple has decided to make four perfect 12-inch pizzas and “fuse” them together on one giant serving platter.

With M5 Pro and M5 Max, the Cupertino crew has gone back to the drawing board and rewrote the rules with chips on a system (CoS). Instead of just jamming it altogether, multiple specialized chips are fused together to act as one — tightly bonded so the software thinks it’s still just one chip.

• It’s much better for efficiency — if one pizza as a hair, you only need to toss that one small piece
• You still get that size advantage — you can still feed the masses a ton of pizza
• The Fusion Architecture magic — the “platter” these pizzas sit on is so fast that the person eating it (the software) can’t even tell they’re separate pizzas

Intel’s been the first big chipmaker to jump into this with Core Ultra Series 3, and not only has it made silicon production much more efficient for them, as you can see from numbers, it’s also brought a big performance/power efficiency boost with it.

And in Apple’s case, you can see how the floor of the M5 Pro chip has been raised, the ceiling of what a MacBook Pro can do has effectively doubled, and the architecture is now future-proofed with simplicity (all they need to do for an eventual “M5 Ultra” is just fuse more blocks onto it).

M5 Max: by the numbers

So how does this fusion architecture perform? We’ve got a solid idea from this Geekbench CPU test score.

That’s a 5.3% increase in single-core and 13% jump in multi-core speeds. Through this lens, it’s a minor jump, but then you take a step back and realize it’s actually faster than the M3 Ultra — Apple’s current workstation-class beast found in the Mac Studio.

Intel Core Ultra Series 3 is similar too, where the CPU speed increases are relatively small. But it’s in the GPU where you’ll see the biggest gains. Just like you saw with Team Blue, Apple’s touting some serious performance increases — all thanks to every single one of those 40 GPU cores having a neural accelerator in them:

• A 50% graphical performance boost
• 4x faster LLM prompting (making Siri and local AI feel instant)
• 3x faster video rendering in DaVinci Resolve (as the GPU is now AI-aware)

Also, shoutout to the 20% improved memory bandwidth to make things a whole lot faster, and the claimed 2x improvement in SSD speeds to 14.5 GB/s.

The modular future

At the beginning of this year, I went to an Intel event where they broke down the architectural shift of Core Ultra Series 3 by giving everyone a Lego set to build the chip. I thought this was a little weird at first (not a problem — love Lego), but the further we get into this year, the more I think it’s the perfect analogy.

We are entering an era of silicon Lego. This whole idea of waiting for a single giant chip is over, because the future is about how well you can glue smaller chips together. For the last few generations, Apple focused on making the “pizza” denser (moving from TSMC’s 4nm process to 3nm in chipmaking).

But we’ve reached a point where the ingredients can’t be made any smaller without things getting weird at a sub-atomic level. So by moving to chips on a system, M5 Pro and Max have broken the oven wall — not limited by how big a single pizza can be, but only by how many they can stick together.

Like I’ve been saying, 2026 is a huge year of change for laptops — with the biggest change happening in the brain of each system. And now with the ceiling virtually eliminated, the sky’s the limit for how far these performance gains can go.

Now, if you excuse me, I’m feeling hungry.

More from Tom's Guide

• MacBook Neo vs MacBook Air M1: There’s a clear (and colorful) winner
• MacBook Neo benchmarks leak — how does it compare to iPhone?
• Did you spot the MacBook Neo's clever message? It's hiding in plain sight

Nvidia GTC 2026 kicks off on March 16 and promises to deliver some big announcements from Team Green. While this is sure to be an AI-heavy show, we’re also expecting important hardware announcements, as CEO Jensen Huang recently teased that his company will unveil "several new chips the world has never seen before."

You may be wondering why you should tune into GTC, since it’s not as consumer-focused as CES. Even though there will be plenty of talk about data centers and AI, Nvidia is likely to drop hints or announce technology that could end up in the devices you use every day, like the best gaming laptops.

We’re covering all things GTC to keep you updated on Nvidia’s plans. Here are some of the biggest announcements we anticipate.

Nvidia’s mystery chip

Given Huang’s comments about a potentially revolutionary AI chip, we can expect a major reveal of a new inference-focused chip, possibly related to the Feynman architecture designed for faster AI tasks. This wouldn’t be a typical GPU, since it’s built for “agentic AI,” where systems act autonomously.

What does this mean for us? Imagine your phone or laptop running AI apps that can edit photos, summarize emails, or plan your day almost in real time. While these features exist, they can feel slow or unreliable. A chip like this could make AI more useful for everyday tasks, and do it all on-device instead of through the cloud. For gamers, it could lead to better upscaling tech, especially for lower- and mid-range PCs and laptops.

The Arm-based N1X CPU

Speaking about chips, there have been rumors about Nvidia’s N1X, an Arm-based CPU for the best computers and best laptops. But with N1X delays reported, GTC might only offer a sneak peek rather than a full launch. If this processor appears, expect discussions on integrated graphics that could challenge Intel’s Panther Lake.

Windows on Arm is now viable, thanks to Qualcomm’s Snapdragon X chips, so an Nvidia Arm chip would be huge. It could create a four-way race between Nvidia, Intel, Qualcomm, and Apple for who can make the fastest, most energy-efficient laptop chip. We might not get much N1X news, but it’s what I’m most excited about as a laptop reviewer.

Next-gen GPU tease

As we recently reported, Nvidia won't release new gaming GPUs in 2026, as they’re focusing on making chips for AI data centers rather than consumer GPUs. However, we might see some architecture teases.

I expect detailed looks at the Rubin GPU, which aims to provide five times the power of Blackwell for AI workloads. Feynman could be a significant leap and might use TSMC’s advanced 1.6nm process.

If you’re a PC gamer, pay close attention to what gets announced, as Nvidia’s data-center tech often influences what runs your PC. For example, Rubin could mean GPUs capable of 8K gaming or ray-traced worlds without overheating your system. Feynman’s efficiency could also improve laptop battery life. Plus, any new tech could lead to an upgraded version of DLSS, making games look and run better.

Physical AI and robotics

I expect Nvidia to also focus on “physical AI,” which combines simulation with real-world robotics. We saw some of this at CES, and I expect the company to expand on it at GTC. Look for sessions on AI factories and multimodal systems that combine vision, language, and action.

It sounds like sci-fi, but if you’ve seen videos of robots doing coordinated dancing, you know we’re getting closer to a Jetsons-like world. Maybe one day we’ll see affordable home robots that can clean, cook, or help those in need, all powered by Nvidia’s Omniverse platform. Nvidia will certainly paint an exciting picture, and I’ll be curious to see if any of it becomes real products.

Agentic AI

You’ll hear about agentic AI, where models go beyond chatbots to multi-agent systems that work together on complex tasks. GTC will likely include workshops on LLM augmentation, prompt engineering, and deploying inference at scale. Essentially, making AI more proactive rather than reactive.

While current LLMs like ChatGPT are useful, they’re not perfect or truly “intelligent.” Agentic AI could deliver genuinely smart AI systems that learn instead of just predict. We’ve heard promises of AI serving as virtual assistants, but these often need a lot of guidance, which defeats the purpose. I remain skeptical about seeing true agentic AI soon, but GTC might start changing my mind.

Outlook

Although Nvidia GTC 2026 will focus heavily on AI, it’s worth tuning in to see what Nvidia is working on and how it could benefit our work and gaming experiences.

Nvidia GTC runs from March 16 to 19, and we’ll be covering it all. The “GTC Live 2026” pregame show starts on March 16 at 11 am ET (8 a.m. PDT), before Jensen Huang delivers his keynote speech at 2 pm ET (11 a.m. PDT). I’ll be watching to admire his leather jacket and hear his insights on Nvidia’s latest plans. Stay tuned for more!

More from Tom's Guide

• It’s been 25 years since Nvidia GeForce 3 — and I think gamers accidentally built the AI era
• Resident Evil Requiem turned me into a path tracing believer — and Nvidia's DLSS 4 made it terrifying
• RAM prices keep going up — what is RAMageddon, and why is it getting worse?

Nvidia GTC 2026 is gearing up to be quite the show, with CEO Jensen Huang now hyping up the annual AI conference with "several new chips the world has never seen before." But what chips can we expect?

During an interview with media site Korean Economic Daily, Huang expressed that GTC 2026, taking place from March 16, will unveil "a chip that will surprise the world." While there's no indication of what this chipset could be, it's easy to think the upcoming Nvidia N1X for PCs is a likely candidate. However, this may not be the case.

Huang is teasing something big from Team Green next month, and while it's looking to be next-gen chips to power future AI data centers, it could be the beginning of the architecture that we'll see in RTX 60-series GPUs.

A mystery chip to 'surprise the world'

During the interview, Huang notes that "nothing is easy because all technologies are at their limits," but with the partnership between Nvidia and SK Hynix, one of the three semiconductor companies, "nothing is impossible."

This only makes Huang's statement of revealing "several new chips the world has never seen before" at GTC 2026 more bold, and while the long-anticipated Nvidia N1X and N1 chips for desktops and laptops are a good contender, looking to take on Intel's Panther Lake CPU lineup (especially in terms of integrated graphics performance), we've heard N1X is rumored to be delayed.

As with past GTC events, Nvidia generally focuses on the expansion of AI-driven software and chipsets, and we've heard about Team Green's new Vera CPUs and Rubin GPU that offers 5x more power than Blackwell to drive the AI data centers of the future at CES 2026.

This indicates we could see its new Rubin GPUs for AI take the spotlight at the show, as they've been in full production as of January 2026, as announced by Huang during CES 2026. However, Nvidia could throw a curveball by revealing its Feynman microarchitecture for next-gen GPUs.

That's considering Huang stating, "this team has worked incredibly hard to meet the great challenge of Vera Rubin and HBM4," referring to SK Hynix and its development of HBM4 (High Bandwidth Memory 4), designed for AI data centers and cutting-edge computing.

This remains to be seen, of course, but if Nvidia is set to unveil a world-surprising chip, then we can expect Team Green to bring the next evolution of implementing AI in the entire tech industry.

A glimpse of RTX 60-series?

With reports of Nvidia not releasing a new gaming GPU this year, this not only pushes back RTX 50 Super GPUs, but also RTX 60-series. And this is due to the RAM crisis. For consumers, this is disappointing, and having a new chip for AI data centers likely being the big surprise at GTC 2026 isn't ideal.

However, consider this. Nvidia Blackwell architecture (what you'll find in the latest RTX 50-series GPUs) was first designed for data centers, and then for consumer GPUs, around a year or two later. With this in mind, with the developments of Rubin now in place, we may see how this architecture comes to Team Green's next-gen graphics cards.

For now, we'll have to wait and see what Huang has up his (leather) sleeve, and it won't be much longer, seeing as Nvidia GTC 2026 is kicking off on March 16. Don't expect all-new RTX GPUs or N1X, but do expect a big reveal that could deliver a taste of Nvidia's consumer graphics cards lineup down the line.

More from Tom's Guide

• AMD just hinted the next-gen Xbox could ‘launch in 2027’ — here’s why I’m not convinced
• We just benchmarked the Intel Core Ultra X9 388H — AMD is officially on notice
• Panther Lake vs Apple M5 benchmarks — 'Intel has done the incredible'

Something interesting happened in AMD’s earnings call, and I’m not talking about the stock price dropping significantly. CEO Lisa Su started talking about its games console chip business, and let slip that “development of Microsoft’s next-gen Xbox featuring an AMD semi-custom SoC is progressing well to support a launch in 2027.”

She also mentioned the Steam Machine is still set for “early this year” too, but it’s that Xbox comment that piqued my interest, as this conflicts with what Microsoft has said in the past. So let’s break down what I predict will be in this next console, and why I’m skeptical about the launch year.

Predicting the next Xbox specs (way too early)

So as Xbox president Sarah Bond said, this is set to be a “very premium, very high-end, curated experience.” With a higher spec ceiling to reach (and a rumored cost of $1,000), that gives Microsoft (and AMD) a lot of room to work in to create what is set to be a console/PC hybrid.

I’m going to go through the rumors I’ve seen and given them a rating out of 10 on likelihood (in my opinion of course):

• AMD Zen 6 chipset architecture (7/10): This next gen CPU architecture is set to rollout in late 2026 — moving to TSMC’s new manufacturing processes to stuff more transistors on there for improved performance. In my mind, it seems likely.
• RDNA 5 GPU (6/10): Xbox Series X sports AMD’s RDNA 2 GPU architecture, which was Team Red’s graphics tech back in 2020. While I can see the logic here (RDNA 2 launched in early 2020 for PCs and came to consoles later that year), given the GPU squeeze at the moment for AI infrastructure, I can see the current-gen RDNA 4 coming instead.
• Up to 48GB of GDDR7 RAM (5/10): Casting serious doubt on this in the current RAM price crisis. Of course, I have my (rather selfish) prayer circle that this AI bubble bursts and components are reasonably priced again. But even with that in mind, I still anticipate 16GB of VRAM.
• AI and neural rendering (10/10): AMD has pivoted hard to AI trickery using FSR Redstone, and the PS5 Pro already does this. I would be shocked if the next-gen Xbox didn’t sport this.

Why I’m pressing ‘X’ to doubt

Well, my suspicions come from what Microsoft itself said. Back in the FTC vs. Microsoft court battle, documents revealed that the next Xbox (and PS6) were set for 2028. This “hybrid game platform” is set to combine local hardware and cloud computing.

Of course, this could’ve changed over the years since this court battle happened, and the “very high-end” comments from Bond does suggest that a pivot is on the table here. But let’s go deeper.

One thing that’s clear is RAMageddon is set to impact console gaming, and with next-gen systems, companies like Microsoft and Sony have a decision to make: Stick to this expected timeline and make it hurt, or wait it out for this DRAM drought to correct.

If I were a betting man, I’d go for the latter. But I’d love to know when you think the PS6 and next-gen Xbox will launch.

More from Tom's Guide

• 'The market environment is indeed challenging': Nintendo president sounds the alarm on 'pressure' caused by surging memory prices as Switch 2 forecast stays at 19 million
• AMD is cooked: Intel has ‘more planned’ for Core Ultra Series 3 in gaming handhelds — XeSS 3 multi-frame generation could be huge
• I tested Intel XeSS 3 vs Nvidia DLSS 4: my laptop tests made me rethink the future of integrated graphics

It's been a long wait for Nvidia's N1X and N1 CPUs, which have recently been confirmed by CEO Jensen Huang as a joint project with MediaTek. But we may be standing by a while longer until we see these chips finally arrive on PCs.

According to reputable leaker Moore's Law is Dead (MLID), Nvidia's N1X CPUs are suffering from bugs and software issues, leading to a delay in when we'll see Arm-based Windows laptops. Apparently, neither Microsoft nor Nvidia is rushing to make sure the SoCs (System on Chips) are fully supported on Windows.

Last year, rumors pointed to the N1X and N1 chips arriving in early 2026, but the recent development indicates that Team Green's CPUs won't launch on laptops until the second quarter of 2026, or even the summer.

Moreover, a previous leak claimed Microsoft's next-gen OS had caused an N1X CPU delay, so there may be ongoing compatibility issues with Team Green's Arm-based chip and Windows that both tech giants need to solve.

For now, this leak may specify why we haven't seen the launch of Nvidia's next-gen chip yet, especially during CES 2026. Initially, it was believed Team Green and MediaTek would announce the CPUs during Computex 2025. Now, it's looking like the reveal may happen during Computex 2026.

Nvidia N1X laptops are coming

While it's still unconfirmed when Nvidia will finally unveil its N1X and N1 chips in upcoming PCs, we at least have an idea of what laptops will come equipped with the all-new CPUs — if leaks are to be believed.

So far, we've heard that Dell's gaming brand Alienware will be among the first to launch new gaming laptops featuring the Nvidia and MediaTek CPU. The chips are expected to deliver the same performance as an RTX 4070-equipped laptop, after all, and we may see Alienware's new ultra-slim gaming laptop come with an N1-series chip.

According to tipster Huang514613 on X via VideoCardz, we should also see Nvidia N1X and N1 SoCs in upcoming Lenovo laptops, with product coding pointing to the launch of the chips. This includes the Lenovo IdeaPad Slim 5, Yoga Pro 7, Yoga 9 2-in-1 and Legion 7.

Not only does this indicate Team Green's N1X and N1 chips are both for laptops (previously, the N1X was indicated to be a CPU for desktops), but it also shows Nvidia's SoCs are for gaming and productivity laptops alike.

Of course, we'll have to wait and see when Nvidia and MediaTek's AI chips arrive, and the PCs we can expect them in. In the meantime, it's looking like we'll have to wait until summer before we see Nvidia enter the CPU market for consumers, and that gives more breathing room for Intel's Core Ultra 3 CPUs.

More from Tom's Guide

• RAM prices are rough right now, but I found 5 DDR5 kits that beat the crisis
• New leak claims Intel Panther Lake laptops could cost up to $2,400 — and my wallet is already crying
• I ditched my regular 'Pro' laptop for a gaming laptop for work — and my productivity just doubled

After months of Nvidia's N1X and N1 chips rumors, CEO Jensen Huang has finally confirmed its Arm-based CPUs designed for AI PCs in collaboration with MediaTek — and it's already taking aim at Intel Core Ultra Series 3 and AMD Ryzen AI 400.

In an interview during Nvidia's year-end party in Taiwan, Huang commented on work with MediaTek for the hugely anticipated next-gen N1X and N1 chips, which have been tipped to power AI-based desktops and laptops, respectively. According to Taiwan media outlet UDN, Huang states these CPUs have been specifically geared for PCs with "powerful AI capabilities."

The report notes that Nvidia's collaboration with MediaTek will offer a System-on-Chip (SoC) with "low power consumption but excellent performance," which recent rumors back up. Leaks indicate that N1X will be closely tied to Team Green's GB10 Superchip found in the DGX Spark, while the N1 for laptops may come in the form of a cut-down version of GB10.

With the joint project with MediaTek confirmed, Nvidia's N1X and N1 SoCs already look to stir up the CPU competition. We've heard that the N1 series will reportedly deliver the same performance as an RTX 4070-equipped laptop, but with improved power efficiency, taking integrated gaming performance in PCs even further.

While Jensen doesn't comment on when we'll see these Arm-based chips in upcoming PCs, reports have suggested an early 2026 launch, with wide availability later in the year. Other rumors believe the N1-series chips were delayed, pushing back the release to late 2026. We've heard that Nvidia's Arm-based SoC may power Dell's Alienware laptops down the line, and it could match the timing for Alienware's new ultra-slim gaming laptop.

However, with Computex 2026 coming up, this may set the stage for Nvidia and MediaTek to reveal their AI-focused CPUs.

Nvidia is coming for the CPU crown

Since benchmarking the Intel Core Ultra X9 388H, Team Blue's flagship laptop processor, we've been hugely impressed by the performance and power efficiency gains the chip delivers. Especially when it comes to Intel XeSS 3 and its integrated graphics for gaming — seeing Cyberpunk 2077 at High settings reach 217 frames per second is a highlight.

Without XeSS and its multi-frame generation or super resolution, Intel's CPU is closer to the performance of an RTX 4050. Still impressive, considering this comes from Arc B390 integrated graphics on the chip, but if rumors ring true and Nvidia's N1-series reaches RTX 4070 levels of performance, that's a significant gap to beat.

Rumors indicate that the N1 chip is expected to use 65W power to match the performance of a 120W RTX 4070 gaming laptop, and another source suggests the chip would offer a TDP (Thermal Design Power) of 80W to 120W. Intel's Core Ultra 3 chips have shown they thrive on low power consumption, too, with the Asus Zenbook Duo we tested hitting 45W.

For now, it's guesswork on what the N1X and N1 CPUs will deliver, but it appears to take the fight to Intel's impressive offering, along with AMD's Ryzen AI 400 series, according to the benchmarks we've seen in the Asus Vivobook 14 we tested.

The CPU wars are heating up this year, and if the ongoing RAM shortage doesn't get in the way, your next laptop or desktop is looking to see a significant upgrade.

More from Tom's Guide

• RAM prices are rough right now, but I found 5 DDR5 kits that beat the crisis
• AMD is cooked: Intel has ‘more planned’ for Core Ultra Series 3 in gaming handhelds — XeSS 3 multi-frame generation could be huge
• 'Google’s leaked Aluminum OS looks like the ChromeOS I've waited years for': why I might actually ditch Windows 11

We just tested Intel Panther Lake, and I’m here to tell you it’s the real deal. After years of Intel playing catch-up, this new architecture feels like a significant leap forward, offering the kind of raw performance and longevity we’ve long wanted from a Windows laptop. It's a night-and-day improvement over the already solid Lunar Lake, and shows that Intel isn't just treading water anymore.

Though Panther Lake has made great strides in performance and efficiency for the best Windows laptops, it has a mighty competitor—namely, Apple’s M5 chip. As I noted in my MacBook Pro M5 review, Apple’s silicon delivers stellar performance and phenomenal battery life. That’s not surprising, given how M-series chips have been the gold standard in those areas, but it makes for a formidable opponent for Intel to tackle.

We’ve officially tested the svelte Samsung Galaxy Book6 Pro with its Intel Core Ultra X7 chip and the dual-screen Asus Zenbook Duo (2026) with the beefier Core Ultra X9. In this piece, I’m going to compare our lab results for these Windows machines against our Apple M5 benchmarks.

Since we're comparing x86 architecture to ARM, I’ll also dive into what this means for app compatibility and your daily workflow. So, is Panther Lake a true M-series rival? Read on to find out!

Performance benchmarks

Let’s start with the raw numbers. I’m looking at our Geekbench 6 CPU scores and our Handbrake video transcoding results. Geekbench measures a chip’s overall "brain power," while our Handbrake test shows how long it takes to convert a 4K video into 1080p.

As you can see, both Panther Lake chips hold their own against the Apple M5 in the multi-core portion of the test. However, the M5 absolutely crushed Intel’s chips in single-core performance.

So, what does that actually mean? Think of single-core performance as having one world-class chef who can whip up a single complex dish really fast. Conversely, multi-core performance is like having a team of cooks working together to finish a big banquet. For daily tasks like browsing the web or opening Word, that high single-core score on the M5 makes everything feel just a bit snappier. But for heavy-duty multitasking, Intel is in the same league.

As with single-core performance, the M5 chip pulled ahead in the Handbrake test, clocking in almost a minute faster than the Core Ultra X9. In the real world, this is a big deal for anyone who does even light video work. If you're exporting a vlog for YouTube or shrinking a movie file for a long flight, the M5 will save you time.

Graphics benchmarks

Next up, let’s look at one of my favorite things: graphics. I’ve listed the results from 3DMark Steel Nomad and Solar Bay. Steel Nomad measures standard gaming performance at 4K, while Solar Bay focuses specifically on Ray Tracing — the tech that makes lighting and reflections in games look realistic. I chose these newer tests over the older Wild Life Unlimited Extreme ones because, frankly, if you’re buying a 2026 laptop, you want to know how it handles modern tech.

The Apple M5 chip scored the lowest here, though its Solar Bay results are still commendable since that test is notoriously heavy on Ray Tracing. Still, Panther Lake has a clear advantage over the M5 when it comes to gaming performance. That's a huge win for an integrated GPU (iGPU) that doesn't rely on a bulky, dedicated graphics card.

What’s really interesting here is that Intel’s Xe3 architecture is finally delivering on the promise of real gaming on a thin-and-light laptop. While the M5 is great for creative work, Windows still has the edge for gaming compatibility.

Seeing the Asus Zenbook Duo and the Galaxy Book6 Pro consistently pull higher frame rates means you can actually play modern titles at decent settings without things turning into a slideshow. Intel has essentially turned these productivity machines into "stealth" gaming rigs.

Battery life

Here’s where things get really fun: battery life! Please note that this comparison isn’t exactly "apples-to-apples" (pun very much intended) since we’re dealing with different hardware configurations. Also, keep in mind that both Panther Lake laptops we tested feature OLED panels, which are gorgeous but notorious battery hogs.

As you can see, the MacBook Pro M5 outlasted its Panther Lake-powered rivals by over three hours! These results might seem unfair, and I’d agree with you if these were typical Windows laptops that lasted for eight hours at best. However, getting nearly 15 hours of battery life from a Windows machine is insanely impressive. It’s a testament to Panther Lake’s endurance and proof that Intel has finally closed the gap that once made MacBooks the only real choice for all-day use.

Compatibility: x86 vs. ARM

To understand these results, you have to look at the engines under the hood. Intel uses x86 architecture, while Apple uses ARM. For years, ARM (originally designed for phones) had the advantage because it was built for efficiency — it stayed cool and sipped battery. x86 was the gas-guzzler: powerful, but hot and power hungry.

The advantage of x86 is its wide compatibility. Every app, game, and legacy program ever made for Windows runs natively on Panther Lake. The disadvantage has always been the heat and battery drain. However, with Panther Lake, Intel has finally figured out how to make an x86 chip act like an ARM chip, giving you that 15-hour battery life without sacrificing your ability to run any software you want.

ARM, on the other hand, is the king of performance per watt. It’s why the M5 can last 18 hours while still crushing single-core benchmarks. The downside? If an app isn't built specifically for ARM, it has to be "translated" (using software like Rosetta 2), which can occasionally cause a performance hit or compatibility glitches. For the average person, Apple’s ecosystem is seamless, but for those who need niche Windows software, the x86 architecture is still the safest bet.

Bottom line

So, is Panther Lake a true rival to Apple's M-series? Absolutely. For the first time in a long time, buying a Windows laptop doesn't feel like a compromise on battery life.

Intel has managed to pull off something incredible here: they’ve delivered a chip that can go toe-to-toe with Apple’s efficiency while still offering the raw graphics power that Windows users crave. Even though Windows 11 can still feel bloated compared to the streamlined macOS, Panther Lake makes the OS feel faster and more responsive than it would with a less powerful processor.

I’m genuinely impressed with how far Intel has come with its "M1 moment." The Samsung Galaxy Book6 Pro and Asus Zenbook Duo are not just awesome Windows laptops; they're great laptops, period. They set a high standard for all the other Panther Lake laptops I’m sure to review in 2026. Color me excited!

That said, Intel isn’t in the clear yet. While Team Blue has shortened the gap with Apple, I’m eager to see how Panther Lake stacks up against the Snapdragon X2 Elite. Qualcomm’s ARM-based Windows chips are the next big rival, and if they can beat Intel on battery while matching them on performance, we might be looking at a three-way silicon war. That’s a story for another day. But for now, Intel is officially back in the game.

More from Tom's Guide

• Forget Nvidia: Integrated graphics could be the future for gaming laptops
• Intel confirms XeSS 3 Multi-frame generation is coming to Arc B580 next month
• I just tried Apple Creator Studio — and I’m never going back to Adobe

The future of handheld gaming is shifting, with the new Intel Panther Lake and its impressive Core Ultra X9 388H CPU leading the charge. For gaming handhelds, this is looking to be a game-changer, and it's already beating AMD where it matters most.

With its next-gen XeSS 3 (Xe Super Sampling) and multi-frame generation tech, Intel has made significant gains for its Arc B390 iGPU — pushing integrated graphics to new highs on laptops. And, after benchmarking the Asus Zenbook Duo with Intel Core Ultra X9, the results are clear.

While seeing Cyberpunk 2077 running at 217 frames per second (FPS) at 1080p with Ultra settings impresses, the TDP (Thermal Design Power) it uses to reach these results will be massively beneficial for the future of gaming handhelds. In fact, it's seeing better performance than AMD's Strix Halo (Ryzen Max AI+) when it comes to low power consumption.

Not only does it outclass the AMD Ryzen Max AI+ in a Framework Desktop in single-core (see the table below), but recent benchmarks from YouTuber The Phawx have shown the Intel Core ultra X9 388H delivers stronger performance at lower TDP.

According to the results, the X9 CPU with its integrated Arc B390 graphics was able to hit 46.6 FPS at just a TDP of 15W in Black Myth: Wukong with 720p resolution at low settings. Compared to the AMD Strix Halo (featuring integrated Radeon 8060S graphics), this reached 39 FPS with the same configuration.

Not only this, but Team Blue's Core Ultra X9 388H even managed to reach 30 FPS at just 10W. Considering the ROG Xbox Ally X with its AMD Ryzen AI Z2 Extreme offers up 16.6 FPS at 35W TDP in Cyberpunk 2077 at 720p and Ultra settings, at 11.9 FPS at 17W, this shows the gains Intel has made with its CPU for handheld gaming.

This supports the benchmarks we've seen in Intel's Core Ultra 3 X9 chip in the Asus Zenbook Duo, with it using 45W as opposed to an Acer Predator Triton 14 AI with RTX 5070 using 140W. Of course, with its discrete GPU, Nvidia's graphics card offers better performance, but with lower TDP, Team Blue's chip still impresses.

That's a promising outlook for upcoming gaming handhelds, and Intel has previously confirmed it will be launching a suite of handheld gaming consoles boasting Intel Core Ultra Series 3 at CES 2026.

"They are the testament of our strategy to bring what the market requires," an Intel spokesperson told Tom's Guide. "A powerful CPU, a powerful GPU, packed with great efficiency within the same solution. With our 12Xe configuration, we deliver performance that approaches discrete GPU levels in thinner, lighter, and more portable devices. This meets the needs of customers and reflects our strategic commitment to rapidly advancing the capabilities of integrated GPUs."

Multi-frame generation is a big win

Aside from improved power efficiency, Intel's XeSS 3 is already knocking it out of the park with its frame generation and resolution scaling.

This would be a major perk for upcoming gaming handhelds, and while Intel confirmed to Tom's Guide that it's coming to the desktop GPU Intel Arc B580 next month (making it the only GPU we'd recommend given the RAM price crisis happening), we asked them about multi-frame generation coming to gaming handhelds.

"Gaming handhelds are an exciting and growing segment where we have more planned, featuring Intel Arc graphics. We’ll share more later this year."

Not much to go on, but considering these chips for gaming handhelds will "feature Intel Arc graphics," in other words, the latest XeSS technology we've seen in the Arc B390, it's looking likely we'll see huge gains in frame generation in gaming handhelds, too.

Combine this with the Core Ultra X9 388H's significant performance and power efficiency gains we've seen, and the future of the best gaming handhelds is looking bright. While AMD claimed its Ryzen AI CPUs beat Intel Panther Lake for gaming and AI, we'll see what Team Red has to offer when it comes to handheld gaming.

More from Tom's Guide

• Forget Nvidia: Integrated graphics could be the future for gaming laptops — here’s why
• I just tested the Samsung Galaxy Book6 Pro — and Panther Lake is the real deal
• ‘Chaotic’ RAM pricing won’t kill PC gaming, a CEO told me — and history backs him up

AMD is taking shots at Intel’s upcoming Panther Lake CPUs. Just ahead of the Intel Core Ultra Series 3’s launch, AMD is showing how its Ryzen AI MAX and Ryzen AI 400/300 mobile CPUs stack up against Panther Lake. As you’d expect, AMD’s infographics paint Team Red’s processors as the superior option.

In the chart below (which we received via email from AMD), you can see that AMD places its CPUs and Intel’s into four categories: Premium, Thin & Light, Mainstream, and Entry. Unsurprisingly, since this is an AMD deck, the company's processors effectively win across the board here for content creation, gaming, and AI.

AMD also takes time to debunk Intel’s claims, saying it is expected to have “the fastest processors for AI PCs.” For instance, the AMD Ryzen AI Max 395+ supposedly has 37% faster graphics and twice the processing threads than Intel’s Core Ultra X9 388H CPU.

AMD also said that Intel’s own data shows “almost no advantage” in performance and efficiency over Lunar Lake (Core Series 2), and that it expects “major processing and graphics wins” with its AMD Ryzen AI 400 series chips.

Since all of this is based on AMD’s internal testing and projections, we shouldn’t take it as gospel. Based on what I’ve heard, it seems that Panther Lake will be the real deal in terms of performance and efficiency. Though the jump from the original Core Ultra to Core Ultra 2 might not have been monumental, Series 3, or Panther Lake, could make Intel’s CPUs on par with the likes of Qualcomm’s Snapdragon X or even Apple’s M-series chips. That said, we also can’t rule out an AMD upset, especially if the company’s claims are true.

With the first Panther Lake laptops launching next week, we should expect reviews to paint a clearer picture. We at Tom’s Guide will conduct our own testing, so be sure to come back here to see what the results are. As always, stay tuned for more!

More from Tom's Guide

• Intel's Panther Lake chip graphics look 50% faster in early benchmarks
• I just got up close with the first Intel Panther Lake laptop
• Intel Panther Lake could power the future of handheld gaming

Asus pulled out all the stops with its lineup of next-gen gaming laptops at CES 2026, with its bonkers dual-screen Asus ROG Zephyrus Duo and a Death Stranding-inspired ROG Flow Z13. But one key device surprisingly didn't make an appearance: the Asus ROG Zephyrus G14 with Panther Lake.

In many ways, this is Asus' flagship machine that's one of the best laptops around, and while there were updated machines with RTX 50-series, the leaked integrated model was canned. Well, sort of, as we got a chance to see the ROG Zephyrus G14 with an Intel Core Ultra Series 3 chip during a hands-on demo in December, and there was no discrete GPU in sight.

Sporting an Intel Core Ultra X9 388H with integrated Intel Arc B390 graphics, this would be the first Zephyrus G14 without a GPU. As with the current Asus ROG Zephyrus G14 (2025), these laptops usually come with graphics cards now up to an Nvidia GeForce RTX 5080. However, with the power to game on an Intel Panther Lake chip, Asus was looking to shift gears.

For now, though, this wasn't meant to be. The Asus ROG Zephyrus G14 with an Intel Core Ultra Series 3 chip mysteriously vanished despite an upcoming reveal, but we now have some insight behind its apparent cancellation.

What happened?

During the lead-up to CES 2026, we got a preview of the ROG Zephyrus G14 with Intel Arc B390 power during an Asus hands-on event in December. Because of this, we were anticipating its big reveal during the CES showcase.

However, at the last-minute, Asus pulled the laptop from the show. It was briefly seen at Intel's demo booth, but was reportedly later removed from the area and wasn't seen again (as told by French outlet Les Numériques).

Many believe the ROG Zephyrus G14 was cancelled or at least postponed due to supply chain issues, with Asus later confirming a "change in strategy" for the laptop. Due to this, it is now looking to be cancelled.

After hearing about the Intel Core Ultra X9 G14's cancellation, we started asking around and found other companies confirming a struggle to nail down supply. So our own Jason England put the question to Intel themselves in an interview that will be published soon.

"A perfect combination of industry-level component shortages, significantly increased demand because of how good the chips are and also ramping up production, you're getting mixed signals depending on where the different partners are," Nish Neelalojanan, Senior Director of Product Management at Intel, commented.

"Good news is there are plenty of X9 designs available. So as a gamer, you will be able to go buy an X9 system from your favorite retailers. There will be availability."

While we will see laptops equipped with an Intel Core Ultra X9 chip for integrated graphics gaming performance, it won't be from the Asus ROG Zephyrus G14 right now. But perhaps this wouldn't have been the best fit after all.

Not the right fit?

During an interview with Asus, I asked whether we'd see integrated graphics on laptops, with Asus' Director of Technical Marketing Sascha Krohn stating: "I would say we're definitely getting into the territory where that becomes a possibility." However, the timing may not be right for the ROG Zephyrus G14.

It's clear that gaming on an Intel Core Ultra Series X9 or X7 processor has seen significant improvements in integrated graphics, being able to play the likes of Battlefield 6 at over 140 FPS at max settings without a discrete GPU. However, it still doesn't reach the heights of an RTX 50-series GPU, and the ROG Zephyrus G14 is known for its top-level performance in a compact design.

Asus may have changed plans to make sure its flagship gaming laptop still offered plenty of power to run the latest PC games at a high level of settings, but as Intel states, demand for the chip would mean a very limited supply for the G14, too.

However, this doesn't take away from the perks of an Intel Panther Lake chip being used in a laptop that isn't just made for gaming, but also for productivity, too. After testing Intel Panther Lake, we've seen major gains in power efficiency, and this should result in machines with gaming performance that have a much longer battery life.

At the very least, we will be seeing an Asus TUF Gaming A14 with an AMD Strix Halo chip, giving us powerful integrated graphics in the company's more budget lineup of laptops.

More from Tom's Guide

• I just went eyes-on with Nvidia G-SYNC Pulsar, and now my own gaming monitor feels obsolete
• 'Basically a photograph interacting with you at 500 frames per second': Nvidia CEO Jensen Huang shares his vision of the future of PC gaming, and how the company plans to tackle GPU pricing crisis
• I just tested Lenovo’s Legion Pro Rollable concept, and now my gaming laptop feels boring

Intel showed off how it raised the bar on integrated graphics in laptops at CES 2026, thanks to the arrival of its Intel Core Ultra Series 3 CPUs. Now, after seeing the latest PC games like Battlefield 6 pushing well past 120 frames per second (FPS) at max settings on a laptop not built for gaming, I'm blown away by the results.

No discrete GPU here, just a chip using Intel Arc B390 graphics (a.k.a. 12 Xᵉ cores) that features everything you'd expect from the latest graphics cards. That includes what Intel calls "Modern Rendering," with XeSS 3 tech featuring ray tracing, AI-based upscaling and multi-frame generation up to x4. If anything, it puts Nvidia's lower-end RTX 50-series GPUs on notice.

With Intel Core Ultra X9 and X7 processors, we're getting a 77% bump in faster graphics and 53% increase in faster AI compared to the previous Lunar Lake chips. That's a significant jump, and the end result is thin and light laptops being able to play the latest big-budget PC games without needing a dedicated graphics card — and at a high level.

I got a chance to try out many of the latest titles during a hands-on demo, and I'm already impressed by the high-level performance and smooth frame rates Intel's next-gen Core Ultra Series 3 chips bring to laptops. One thing's for sure: your next gaming laptop may not need a dedicated GPU after all.

A major leap for integrated graphics

As Intel boldly claimed during its big keynote, the Arc B390 in Core Ultra X9 and X7 chips will see Battlefield 6 at Overkill settings (the highest visual settings) and Super Resolution running at 59 FPS. Turn on its multi-frame generation tech, though, and this can be upscaled to a super-smooth 145 FPS.

That was fully on show during a brief demo I tried. Now, Battlefield 6 was already a well-optimized game when I tested it on an RTX 5070 gaming PC, where I was getting over 200 FPS at 4K with DLSS 4 switched on. However, this doesn't take away from how well a laptop sporting a Core Ultra Series 3 chip can handle it, as I was seeing detailed visuals and buttery-smooth framerates as explosions and gunfire went off all around me.

But there was more on show. I also got to try Dying Light: The Beast, where I saw framerates reaching 196 FPS at what appeared to be high settings. As expected, it was stunning running around, dropkicking zombies with beautiful visuals thanks to ray tracing.

Also on show was Clair Obscur: Expedition 33, Marvel Rivals and Painkiller, and while I couldn't see their frame rates or settings, Intel has confirmed that we can expect over 100 FPS with 1080p at high settings using XᵉSS Super Resolution.

Not only does it make playing competitive shooters on an Intel Core Ultra X9/X7-based laptop a breeze, but these new chips can also offer upscaled visuals in demanding PC titles — and that x4 multi-frame gen tech certainly helps.

Outlook

If there's one laptop I'm keeping an eye out for in 2026, it's one boasting an Intel Core Ultra X9 or X7 processor. Team Blue has pushed integrated graphics on its next-gen chip even further with Intel Arc B390, and from the games I've played, it's looking to kick off a new era of PC gaming on thin and light laptops.

Of course, we'll have to put these Intel-based machines through their paces in our testing labs once we get our hands on them, but it's looking like we could see these laptops creep into our list of the best gaming laptops — especially for gamers on the move (and if the price is right, no less).

Speaking of, with Intel also announcing an entire handheld gaming platform with Panther Lake chips, gamers will have a lot to look forward to this year.

More from Tom's Guide

• Dell XPS is back! I went hands-on with the new XPS 14 — and it's the best kind of apology
• I just saw Alienware's new ultra-slim gaming laptop at CES 2026 — and it's nearly as thin as a MacBook Air
• Qualcomm announces Snapdragon X2 Plus CPUs at CES 2026 — and it promises an ‘epic’ upgrade

2026 is the year of the laptop, and the CPU wars are critical to that. I’ve tested Intel’s Core Ultra 300 Series chips and Qualcomm Snapdragon X2 Elite, but there’s one more player in the Windows laptop space.

AMD is back and the rumors are true — Ryzen AI 400 Series chips (codenamed “Gorgon Point”) have just been made official at CES 2026. Here’s everything you need to know about the new silicon from Team Red.

AMD Ryzen AI 400 Series Processor specs

The same, but much better

With Ryzen AI 300 Series chips (codenamed “Strix Point”), AMD tore up the rulebook and overhauled its architecture to point their laptop chips at zippier performance, better local AI, improved graphics and power efficiency.

And now, Ryzen AI 400 Series (“Gorgon Point”) looks set to take that to a new level with a suite of chips that you’ll find in budget, mid-range and premium notebooks that focus on three key things — faster multitasking, better graphics and AI, and improved power efficiency.

Starting with the first, that CPU boost clock has been increased up to 5.2 GHz and RAM speed support has been upped to 8,533 Mega Transfers per Second (MT/s). So faster CPU cores and zippier RAM.

Second, those RDNA 3.5 GPU cores now run at a faster boost clock of up to 3.1 GHz, and the next-gen NPU on the chip die contains up to 60 TOPs of AI performance. And finally, the low power architecture on the chip has been optimized to allow for not just faster unplugged performance, but improved power efficiency too.

By the numbers

Much like everyone else reporting on these chips, we won’t have specific numbers until we are able to give it a full test in our labs. However, there have been several leaks — all of which I’m keeping an eye on in my CPU wars report.

This can give us context as to whether AMD’s numbers can be backed up, and when we take our Geekbench tests of laptops we reviewed over the past 18 months and compare, Team Red’s claims are legit.

But of course, this is looking at it through a vacuum of comparing old to new. Seeing the numbers when compared to the competition of 2026 shows that it’s neck and neck with Intel, whereas the Arm-based likes of Snapdragon X2 Elite have a considerable lead.

That being said, there’s compatibility on AMD’s side here running on x86 architecture and not needing an emulation layer. For tests on the likes of the company’s battery life, gaming and NPU claims, we’ll have to wait and see when we get to properly test these.

Outlook

Geekbench is only half the story here when it comes to chips, but an effective part in showing that AMD has stepped up its game performance-wise for the next generation of laptops and mini PCs coming this year.

But it's also the attention paid to tackling the issues with x86 that make people go for the Arm likes of a MacBook that could pay off for them in the long run — faster performance when unplugged, speedier content creation workloads and multitasking, and of course that overhauled low power architecture for multi-day battery life.

Whether these claims will be backed up, we’ll see when laptops from the likes of Acer, Asus, Dell and more launch over the next few months. But signs point to optimism.

More from Tom's Guide

• I went hands-on with the redesigned MSI Stealth 16 AI+ — it’s actually stealthy now, and gunning for the MacBook Pro
• Gaming on Snapdragon X Elite laptops just got a whole lot better — I tested the latest update and saw 40% faster gameplay
• Intel Panther Lake vs Snapdragon X2 Elite vs AMD Gorgon Point — one chip is already pulling ahead in 2026's CPU war

So far, my team and I have been blown away by the AMD Ryzen AI Max+ 395 (codenamed Strix Halo) — packing integrated graphics so powerful they match up to dedicated Nvidia laptop GPUs. In fact, it’s one of the more crucial chips to 2026 being the year of the laptop.

And in response, Team Red is bringing this prowess to more mid-range options with two new Strix Halo chips. These reduce the total CPU cores and the clock speed, while giving you that same 40-core GPU, and I’m so excited to see these capabilities come at better prices.

Meet the AMD Ryzen AI Max+ family

A gaming powerhouse

Over the course of 2025, one thing became abundantly clear — that AMD Radeon 8060S GPU built into the Ryzen AI Max+ chip is the key reason why “integrated” is no longer a dirty word in PC gaming.

Whether it’s the Asus ROG Flow Z13 or the Framework Desktop, we’ve been stunned at just what that RDNA 3.5 architecture can pull off in AAA games running at buttery smooth framerates. Turn on FSR and that potential gets even higher.

And this year, by compromising a little on the CPU side, that same fully-loaded GPU can come to cheaper systems. Price was always one of the main obstacles here, so it’s great to see the company tackle this head on.

A true workhorse

And one of the key things that grew in the background of all this is the ability to support huge offline large language models via this chipset’s GPU cores. In fact, it can run ChatGPT-OSS 120B 1.7x faster than Nvidia’s own DGX Spark AI mini supercomputer (according to AMD’s own testing).

In a more apples-to-apples comparison, it has 1.4x faster AI performance than an M5 MacBook Pro, and 1.8x faster content creation performance to boot.

Put simply, Strix Halo drove a lot more of the AI PC era than any chip’s NPU ever did, in terms of actually getting stuff done. And I’m excited to see the systems this chip ends up in from the likes of Asus, Acer, HP, Lenovo and more.

More from Tom's Guide

• AMD Ryzen AI 400 Series might finally make charging anxiety a thing of the past — 1.7x higher unplugged performance and big multitasking upgrades
• Gaming on Snapdragon X Elite laptops just got a whole lot better — I tested the latest update and saw 40% faster gameplay
• Intel Panther Lake vs Snapdragon X2 Elite vs AMD Gorgon Point — one chip is already pulling ahead in 2026's CPU war