I was slightly nervous that Nvidia was bringing AI, data center and robotics news to CES 2026, so consider this a huge sigh of relief to see Team Green bring DLSS 4.5 — targeting 4K path-traced gaming at a whopping 240 FPS.
Follow all of our CES 2026 coverage for the biggest gadget news straight from Las Vegas. And be sure to follow Tom's Guide on TikTok for the coolest videos from the show.
For those not in the know, Deep Learning Super Sampling (DLSS) is Nvidia’s AI-driven learning of how a game works and runs, which is applied to making that game run so much better on your gaming desktop or laptop.
So far, over 250 games support it, and over 80% of 2025’s top 20 new games support it, so let’s see what new features are coming with this sizable upgrade.
How DLSS works
Just like other versions of Nvidia’s AI gaming trickery in the past, DLSS 4.5 is split into two things — super resolution and frame generation, which both use those on-board GPU cores for AI smarts in rendering and running games.
The first one takes many hours of gameplay and trains a Transformer model, so the game can be rendered at a lower resolution and visually upscaled by AI — drastically easing the rendering pressure put on your GPU. And the second uses that same model trained on games to add additional AI-generated frames for every rendered frame.
What’s the end result? A bit of a PC gaming breakthrough for mid-range and lower-end cards. As I saw testing the RTX 5060 Ti, the previous version DLSS 4 enabled me to run AAA titles at 4K 120 FPS with the greatest of ease.
And now, this update may sound iterative on paper, but looks set to bring significant gains to the table.
Super Resolution gets more super
Previously, Nvidia’s DLSS ran on a Convolutional Neural Network (CNN) — it did the job, but wasn’t the most efficient, as each pipeline of work is super linear in a network like this. However, flipping to a Transformer Model (like what the likes of ChatGPT and Gemini use) for DLSS 4 brought a breakthrough.
Now it could learn a lot faster and more efficiently, and the end result was a tidy boost in graphical fidelity in super resolution. DLSS 4.5 takes this one step further by doing two key things with the hundreds of games the model is trained on:
- A more efficient training loop: every element from motion and geometry to lighting, motion and game engine data is analyzed and paired with a high-fidelity dataset to get a deeper understanding of the on-screen action. There are even more fail state signals, too, just in case DLSS 4.5 gets something wrong, so it can learn much more rapidly to fix problems going forward.
- 2nd Generation Transformer: With 5x more compute power over DLSS 4, smarter pixel sampling and greater context awareness, this makes all the above possible.
The end result is better picture stability, vastly reduced ghosting, less shimmering on brightly lit surfaces and much smoother edges.
Multi-frame generation gets smarter
And sure, there is a blunt instrument in the form of 6x multi-frame generation — the improved frame pacing also means that in testing, we saw this add only a few extra milliseconds of latency for adding five AI-generated frames for every one rendered picture. Black Myth: Wukong, for example, was found running at over 240 FPS with 55ms response time at 6x over 50ms for 4x.
But most interesting to me is what Nvidia is calling Dynamic Multi-Frame Generation. Instead of slapping on a multiplier and calling it a day — maybe trying to pump too many frames through your monitor’s refresh rate and creating a slight disconnect your eyes will notice — the model can instead be fluid with the various multipliers in real-time to maximize the FPS to your monitor refresh rate.
For example, if you have a 4K 240Hz monitor like the Alienware 2725Q I have, I can tell DLSS 4.5 the refresh rate, and it’ll act accordingly to give me bang on or as close to 240 FPS gameplay as possible.
Outlook
I’ll be honest — while I empathize with the criticisms I see and talk to friends in the industry about (such as the AI generative side of things feeling like a shortcut to not increasing VRAM capacity or core count, and latency complaints), I speak to the regular folks in my life and believe it’s all been blown a bit out of proportion.
Like, I get their frustration, but to most gamers, DLSS unlocks a lot of value from your pricey GPU. It may not be huge news like the RTX 50 SUPER Series, but DLSS 4.5 is another step to stay in the lead when it comes to AI-infused PC gaming.
Of course, there’s much for us to test — pixel peeping Super Resolution and checking out any potential latency issues in Dynamic Frame Generation. So keep it locked on Tom’s Guide for my full testing of these new features before they launch (later this month for Super Resolution and in spring 2026 for the new frame generation tech).
Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds.
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
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.
