Nvidia announced DLSS 5 on Monday, which was swiftly followed by immediate backlash from gamers and developers alike. And while Nvidia CEO Jensen Huang has claimed people are wrong about their distaste for the fledgling technology, Team Green has released some comments that have clarified what it does – for better or worse.
In a statement to YouTuber Daniel Owen, Jacob Freeman, GeForce Evangelist for Nvidia clarified some of the finer points about how DLSS 5 actually works. And, just like Death Stranding 2 animator Mike York suggested, it seems like the algorithm is just taking frames from the game, along with motion vector data and drawing a new image that’s pasted on top.
When Owen asked whether or not the model is just taking a rendered frame as input, Freeman responded “Yes, DLSS 5 takes a 2D frame plus motion vectors as input.” Then, he clarified “DLSS 5 is trained end to end to understand complex scene semantics such as characters, hair, fabric and translucent skin, along with environmental lighting conditions like front-lit, back-lit or overcast, all by analyzing a single frame.”
Nvidia has made the claim about “complex scene semantics” before, but because the model is only taking the rendered frame and motion data as input, it has no way to know definitively the values given to various in-game objects by the developers.
Owen was told the same thing I’ve been told by Nvidia, too, that developers will have “detailed controls such as intensity and color grading. Artists can use these controls to adjust global contrast, saturation, and gamma, and determine where and how enhancements are applied to maintain the game’s unique aesthetic.” But, given what Nvidia has shared about how the model works, these controls seem limited to the kind of sliders you’d find in Adobe Lightroom.
I wonder if the UI will look like Lightroom, too.
What Is Going On Here?
For almost a decade now, starting with ray tracing’s debut on the RTX 2080, Nvidia has been on this quest to maximise “realism” in video games, and that obviously hasn’t changed here. Team Green seems obsessed with making games look as much like its admittedly impressive tech demos as possible. DLSS 5 seems to just be the latest step in that quest.
But while some of what DLSS 5 is doing is impressive, it smashes through artistic intent in order to get there. Sure, the lighting is better in some of the samples Nvidia has presented, but is that added detail worth it if every character gets passed through a yassification filter? I don’t think it is.
Ray tracing works so well with a game’s aesthetic because it’s worked into game engines by necessity. That means that it only does what the art team wants it do, because it’s baked into the engine. From what it seems, DLSS 5 looks to do a rush job on game beautification, only taking a finished frame and essentially running it through ChatGPT to make it look more realistic.
Nvidia has argued against this, saying that the game’s geometry isn’t being changed by the algorithm, but that doesn’t matter. Leaving the source alone doesn’t mean anything if you’re just pasting a new image on top of it.
I have no doubt that as DLSS 5 matures over the next few months it’s going to get more accurate, and make less drastic changes to the original aesthetic, but unless Nvidia goes back to the drawing board with how its implemented, it’s never going to be perfect. It’s always going to make some kind of mistake that’s going to screw with the overall aesthetic.
The Future of DLSS
From the beginning, I’ve thought that DLSS has been a positive force for PC gaming, even if its had a couple of wrong turns here and there. Originally, I thought it would be a great way for folks with less-powerful hardware to be able to run games that would have been shut off to them in the past.
But over the last few years, at least since the RTX 4090 came out in 2022, it has seemed like the opposite has been true. Instead of trying to expand what was possible on lower-end hardware, features like frame generation have instead been about reaching incredibly high frame rates on high-end monitors. Hell, Nvidia itself doesn’t even recommend turning on frame generation unless you’re already getting excellent performance.
DLSS 5 is just the next step in DLSS’s spiral into decadence, and it wasn’t surprising that the new feature needed two RTX 5090s to run. Because, instead of focusing on maximising performance, this is just another feature that’s going to run best on the most expensive graphics cards, while the prices will inevitably rise on less powerful GPUs.
Sure, Nvidia will continue to train the model over the next few months, and it will get fast enough to run on a single GPU. But, you better believe that you’re going to need to have a boatload of VRAM to run it well – and it’s not like that stuff grows on trees right now.
Jackie Thomas is the Hardware and Buying Guides Editor at IGN and the PC components queen. You can follow her @Jackiecobra
