DLSS 4.5 is now live — I tested Nvidia’s upscaler to see which model you should actually use
Is DLSS 4.5 ready for primetime?
Here at Tom’s Guide our expert editors are committed to bringing you the best news, reviews and guides to help you stay informed and ahead of the curve!
You are now subscribed
Your newsletter sign-up was successful
Want to add more newsletters?
Daily (Mon-Sun)
Tom's Guide Daily
Sign up to get the latest updates on all of your favorite content! From cutting-edge tech news and the hottest streaming buzz to unbeatable deals on the best products and in-depth reviews, we’ve got you covered.
Weekly on Thursday
Tom's AI Guide
Be AI savvy with your weekly newsletter summing up all the biggest AI news you need to know. Plus, analysis from our AI editor and tips on how to use the latest AI tools!
Weekly on Friday
Tom's iGuide
Unlock the vast world of Apple news straight to your inbox. With coverage on everything from exciting product launches to essential software updates, this is your go-to source for the latest updates on all the best Apple content.
Weekly on Monday
Tom's Streaming Guide
Our weekly newsletter is expertly crafted to immerse you in the world of streaming. Stay updated on the latest releases and our top recommendations across your favorite streaming platforms.
Join the club
Get full access to premium articles, exclusive features and a growing list of member rewards.
DLSS 4.5 is out of beta and available to use by everyone. Make sure you update your Nvidia app and GPU drivers and it’s all yours across all the games that already support DLSS 4! Just open the app to update or download it manually here.
But after testing it over the past few days since going eyes-on for the first time at CES 2026, I’ll say it can get confusing to figure it all out. So with that in mind, I want to share the results of all my testing and answer two key questions:
- When should I use DLSS 4.5?
- What is the difference between Model L and Model M?
But first, an explainer.
What is DLSS?
Deep Learning Super Sampling is one of Nvidia’s showcase pieces of AI trickery that extracts better performance from your games.
To do so, it will render a game on the GPU at a lower resolution and use a neural network trained on the game to upscale to a higher resolution.
While from the get-go, you can use an automatic mode in most games, which chooses what’s best based on the specs of your PC, there are four modes to pick from:
- Quality: This renders the game closest to native resolution (normally around 66-70% of resolution).
- Balanced: Finding the mix between getting the best possible textures and frame rates, this will render the game at 58% of your targeted resolution.
- Performance: One step down to maximize frame rate, this will often go for 50% of the resolution.
- Ultra performance: And if you want to go all out on frames, ultra performance will target 33.3% of the total resolution being rendered by the GPU.
These can come with a penalty, and these are usually ghosting (faint outlines around moving objects as DLSS fights to keep up with what’s happening on-screen), shimmering (brightly lit objects being a little bit blown out), overdone AI sharpness at times and rendered detail in the background not being clear until you get closer to it.
Get instant access to breaking news, the hottest reviews, great deals and helpful tips.
In terms of model-naming:
- Model K = DLSS 4
- Model L = DLSS 4.5 targeted at ultra performance mode
- Model M = DLSS 4.5 targeted at performance mode and above
When should I use DLSS 4.5?
Put simply: use it if you’re opting for performance modes. Nvidia has been clear in saying that’s what this model is targeting, and when put through testing at higher quality modes like balanced or quality, you’re not getting any noticeable uplift in image quality but a lower frame rate. By the way, all these tests are done with frame generation turned off.


System | Black Myth Wukong (DLSS 4 Balanced) | Black Myth Wukong (DLSS 4.5 Model M) | Cyberpunk 2077 (DLSS 4 Balanced) | Cyberpunk 2077 (DLSS 4.5 Model M) |
|---|---|---|---|---|
RTX 5080 PC | 53 FPS (4K Cinematic) | 53 FPS (4K Cinematic) | 84.53 FPS (4K Ray-tracing overdrive) | 68.96 FPS (4K Ray-tracing overdrive) |
RTX 5070 Laptop | 52 FPS (1200p High) | 44 FPS (1200p High) | 77.84 FPS (1080p Ray-tracing low) | 70.17 FPS (1080p Ray-tracing low) |
RTX 5060 Ti (16GB) PC | 47 FPS (1440p Cinematic) | 45 FPS (Cinematic) | n/a | n/a |
Meanwhile, if you do move to performance, you only get taxed a little, but the image is much more stable with reduced ghosting around objects, better distant detail and less shimmering.


Cyberpunk 2077 settings | DLSS 4 | DLSS 4.5 |
4K Ray-tracing overdrive DLSS performance mode | 86.61 FPS | 84.53 FPS |
4K Ray-tracing overdrive DLSS ultra-performance mode | 126.22 FPS | 125.75 FPS |
Model L vs Model M


In my testing across several games (and after speaking to Nvidia), here’s how to choose between the models:
- Model M: This is the model that’s been trained more for clarity, so this one is recommended for performance mode — driving for more detailed graphics.
- Model L: This is better for ultra performance. You still get taxed in image quality, but nowhere near as much as you see in DLSS 4.
You can see as much in Cyberpunk 2077. The differences are subtle, but in the sharpness of the lamps and the heatwaves coming up from the food cart and distorting the bottles, you can see a little more ghosting and jagged edges in Model L over Model M.



My recommendation to you is that if you are focused on ultimate fidelity, stick with Model K (DLSS 4). Model L is mightily impressive, but DLSS 4.5 Model M is the better all-rounder that Team Green has confirmed will improve over time into quality and balanced modes. But that transition is still in progress to my eyes.
My next steps
So where will my testing take me next? Well, first off, I’ve been hearing that the more advanced acceleration parts of this new transformer model have a bigger penalty on GPUs older than the RTX 50 series. I’m in the process of calling in some older cards and gaming laptops to put this to the test.
But let’s get weird, too. I’ve tested the features you’ll get out of the box, but if you know anything about me, I’m a bit rogue when it comes to testing — pushing the theory behind new tech to its extreme.
In the behind-closed-doors Q&A, Nvidia CEO Jensen Huang talked about how “more and more computation” could be done on “fewer and fewer pixels.” Put simply, DLSS on steroids is where Huang sees all of this heading. I’m setting up what’s needed (DLSS Tweaks) to put this theory to the test — seeing how it looks to render a game at 360p and upscale to 4K.
I’ve seen some of my friends do this, and the results are…well, they’re a mess if you want to play using the settings. But when you think about what’s happening (a neural network analyzing the picture and upscaling everything), it’s a mindblowing preview of the progress being made towards this vision.
Watch this space. I’ll be back with more.
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 just went eyes-on with Nvidia G-SYNC Pulsar, and now my own gaming monitor feels obsolete
- I just tested an 8K VR headset that weighs less than my phone, and it could change VR gaming
- I tried HyperX’s brain-reading headset, and it mindfully made me a better gamer by overclocking my mind

Jason brings a decade of tech and gaming journalism experience to his role as a Managing Editor of Computing at Tom's Guide. He has previously written for Laptop Mag, Tom's Hardware, Kotaku, Stuff and BBC Science Focus. In his spare time, you'll find Jason looking for good dogs to pet or thinking about eating pizza if he isn't already.
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.










