AMD FSR 2.0 Quality & Performance Review - The DLSS Killer 198

AMD FSR 2.0 Quality & Performance Review - The DLSS Killer



AMD has achieved the unthinkable—the new FidelityFX Super Resolution FSR 2.0 looks amazing, just as good as DLSS 2.0, actually DLSS 2.3 (in Deathloop). Sometimes even slightly better, sometimes slightly worse, but overall, this is a huge win for AMD. Take a look at our comparison images—there's a huge improvement when comparing FSR 1.0 to FSR 2.0. The comparison to "Native" or "Native+TAA" also always looks worse than FSR 2.0, which is somewhat expected. When comparing "DLSS Quality" against "FSR 2.0 Quality," spotting minor differences is possible, but for every case I found, I'd say it's impossible to declare one output better than the other; it's pretty much just personal preference, or not even that.

Things look a bit different at the lower end of the spectrum, comparing "DLSS Performance" to "FSR 2.0 Performance." I would say that DLSS is slightly better here, especially when it comes to textured surfaces. Thin geometry also looks a little bit more detailed with DLSS, but it's a very close outcome overall, especially considering NVIDIA has had a lot of time to fine-tune DLSS, whereas FSR 2.0 is on its first iteration. If you actively hunt for rendering issues, you can find them in both upscaling implementations, but I'm happy to report that ghosting, while slightly visible, isn't a serious issue in FSR 2.0. The only noteworthy case is when fine line geometry, like a fence, is sitting behind other fine geometry, like vegetation. DLSS handles ghosting a little bit better overall even though the visual artifacts in DLSS are slightly more distracting than in FSR 2.0.

We also have comparisons with NVIDIA's DLAA (deep-learning anti-aliasing), which renders at native resolution and uses DLSS only for anti-aliasing. This is basically native quality with enhancements. FSR doesn't have any counterpart at this time, but I'm sure if demand is high enough, they'll add this capability. From a technical perspective, it's not that challenging to add on top of FSR 2.0. Not sure if the performance hit of native resolution rendering is worth the tiny improvement in image quality.. maybe for those single-player games that run 200+ FPS anyway, where you want the absolute best in image quality.

With FSR 1.0, AMD pioneered the addition of a sharpening filter to the upscaling pipeline. While this has been possible with NVIDIA sharpening through their Control Panel, too, it wasn't as nicely integrated. NVIDIA has since added a sharpening pass to DLSS 2.0, too, but that option is not available in Deathloop. While sharpening on FSR 1.0 was tied into the "Ultra Quality" preset exclusively, it has been decoupled with FSR 2.0 and is now a separate option with fine-grained control over the amount of applied sharpening. I hope there's a strong recommendation for all developers to expose these settings to the end-user because a selectable range plus "off" is a must for any sharpening filter due to personal preferences. That's also why I separated the "FSR 2.0 Quality" and "FSR 2.0 Quality with "Sharpening" comparison images—lots of people find sharpened images very distracting and prefer the slightly smoother look.

Another novelty is that FSR 2.0 supports dynamic resolution scaling, which works very well. You have plenty of options to play with, including setting a FPS target and a quality minimum, and the algorithms take care of the rest. The way the game switches between resolutions is so smooth I couldn't even spot it—I tried.

In terms of performance, FSR 2.0 deserves praise, too. While it is a little bit more demanding than FSR 1.0, which is not surprising given the additional logic, it's still mighty fast and pretty much identical to DLSS 2.0 on even NVIDIA hardware, which is able to offload many DLSS upscaling operations to the Tensor Cores. No doubt, on AMD hardware, there will be additional optimizations, but we wouldn't be able to compare performance against DLSS then because it's an NVIDIA exclusive technology. On the other hand, you're able to use FSR even on older NVIDIA hardware that doesn't support DLSS (Pascal or even Turing 16-series). I really have to applaud AMD for democratizing upscaling without additional hardware requirements. All we need now is widespread game developer support.

Developer support will be the biggest challenge for AMD I'm sure. While implementing FSR 1.0 is trivial—it's basically an additional shader pass—FSR 2.0 has more serious requirements that are basically identical to DLSS 2.0. For developers that already support DLSS 2.0, adding FSR 2.0 support will be easy, AMD talks about days. The Motion vectors are the biggest challenge here. As a developer, you have to ensure every single object that moves is accounted for or there will be rendering issues. This is still happening with DLSS today; for example, near doors, lifts, fans and other animated objects, 100% the developer's fault. Over time, as companies author their assets with upscaling in mind, this will become a non-issue, though. Another obstacle for AMD will be that NVIDIA has deep pockets and is certainly able to provide various means of "support" to game developers that embrace DLSS. This goes both ways, of course, like we've seen with FreeSync overtaking G-SYNC, and AMD's strategy to partner with game developers for optimized ray tracing effects seems to be working fine. In a perfect world, game developers would be adding support for both DLSS and FSR, which will ensure both companies have to compete and won't stop innovating—who knows what amazing upscalers that will lead to.
Discuss(198 Comments)
View as single page
Jun 28th, 2022 08:43 EDT change timezone

New Forum Posts

Popular Reviews

Controversial News Posts