Dont take Nvidias word for it — try Nvidias awesome new image comparison tool, instead – The Verge

Nvidia wants you to know that AMD’s FidelityFX Super Resolution (FSR) is a mere pretender, nothing like the Nvidia Deep Learning Super Sampling (DLSS) technique its GPUs can use to boost the framerate and/or image quality of your games. To prove it, Nvidia is trying an interesting new tactic by releasing a remarkably powerful image comparison tool called Nvidia ICAT, for free, so that anyone can see for themselves.

ICAT, short for “Image Comparison And Analysis,” is a very simple program that lets you view a stack of gameplay screenshots and even recorded videos side-by-side, automatically lining up the frames, and allowing you to fluidly pan across each scene, zoom in and out, on all of those images or videos simultaneously. It’s got a drag-and-drop interface that makes pixel peeping child’s play, and as known pixel peepers, you’d better believe The Verge is going to be using it for more than just gaming graphics!

Nvidia’s new ICAT tool in action. There’s also an image slider mode.
Screenshot by Sean Hollister / The Verge

But Nvidia’s hope, again, is to show how poor AMD’s FSR is compared to DLSS — or even Nvidia’s own, separate spatial upscaling technique.

In case you need a brief primer: Spatial upscaling techniques, like AMD FSR, run each of your game’s video frames one at a time through a fixed algorithm — and they don’t require a special GPU to run. Nvidia’s DLSS, on the other hand, is a temporal upscaling technique that compares multiple frames and takes account of how things are moving in a video game scene, and processes all that using a neural network that runs exclusively on the Tensor cores you can only find in an Nvidia RTX GPU.

Nvidia illustrates how temporal and spatial upscaling differ.

But before you say “that’s great Nvidia, but I can’t actually buy your new GPUs!” you should probably know that Nvidia has its own spatial upscaler as well, dubbed Nvidia Image Scaling. It’s apparently been buried in the Nvidia Control Panel for some time now, and today Nvidia is open-sourcing it on GitHub with its own SDK and support for every brand of GPU. Developers can natively integrate it into their games if they want, just like AMD’s FSR.

And, the company’s baking Nvidia Image Scaling into its GeForce Experience app so you can turn it on for any game, and adjust an in-game sharpness slider (so you can see the difference) using Nvidia’s overlay.

Is it as good as Nvidia’s DLSS? Not even close, and Nvidia’s the first to admit it. In a briefing with The Verge, product manager Henry Lin hammered us with example after example of how DLSS (particularly the new DLSS 2.3, which suffers from less ghosting around moving objects) not only beats the pants off of Nvidia’s own spatial upscaler, but sometimes performs better than a native image. Particularly when you’re looking at thin objects with lots of edges that tend to flicker:

The communications dish is crisper and less aliased with DLSS.

The pictured communications dish is one place DLSS has an advantage.
Image: Nvidia

Personally, I’ve sworn by native resolution all the way — but Nvidia’s ICAT made me curious enough to put it to the test. So I fired up Deathloop and Back 4 Blood, two games that offer both AMD FSR and Nvidia DLSS, on a 4K screen at their highest quality modes.

In Back 4 Blood, I felt vindicated. Every part of the image looked fuller, crisper, and higher resolution at native 4K, up to and including the roof-mounted antenna:

You’ll want to tap to zoom and/or download, this is a 4K image.
Screenshot by Sean Hollister / The Verge

There were even parts of the scene where AMD FSR clearly beat Nvidia DLSS, like all of these wooden textures:

Maybe Nvidia’s neural network needs more experience with wood.
Screenshot by Sean Hollister / The Verge

But when I tried out Deathloop, it was the opposite story: almost everywhere I looked, DLSS was resolving and preserving details I couldn’t even see at native resolution, and with far, far less flicker from distant aliased objects.

In Deathloop, the VW bus in the background looks decidedly less cartoony with DLSS, and you can see far more detail in distant textures.
Screenshot by Sean Hollister / The Verge

Native 4K, blown up to show detail.
GIF by Sean Hollister / The Verge

DLSS Quality, at the same size. Less flicker, more detail.
GIF by Sean Hollister / The Verge

That said, I still saw moments in Deathloop where DLSS does weird things, too. Scroll up to the top of this post for one example: you can see how DLSS totally changed the sparkly finish of the Strelak Verso pistols and muddies the texture of the dirt ground in the exact same scene as my other screenshots and GIFs.

I’m not a believer in DLSS quite yet, but I’m intrigued. Nvidia’s promise is that, with enough learning, games can actually look better and run faster than native with DLSS turned on. For now, it’s something developers have to enable on a per-game basis, and not all games have the same version to offer, but that could change down the road. Nvidia’s Lin tells me that Cyberpunk 2077 already features over-the-air DLSS updates — when you open the game, it can check with the Nvidia driver to see if there’s a new version and hot swap it at launch.

“Wherever you see DLSS not match native, that’s what we’ll be working on,” says Lin.

This entry was posted in Uncategorized. Bookmark the permalink.

Leave a Reply

Your email address will not be published. Required fields are marked *