What is AMD's FSR, and does it make Nvidia's DLSS obsolete?
Let's analyze these two at a higher resolution.
Much of the hype leading up to the launch of AMD’s FidelityFx Super Resolution (FSR) was built on an assumption that FSR would be a true competitor to Nvidia’s Deep Learning Super Sampling (DLSS).
To the layperson, this is a logical assumption to make. Both technologies lower the resolution a game is rendered at and then use advanced techniques to increase the perceived resolution of the image shown to the player. This frees up processing power and yields higher frame rates for the player, while also allowing the use of hardware accelerated ray tracing without a dramatic hit to performance.
But there are some core differences in how each technology operates, making DLSS and FSR less competing technologies than a convergent evolution to the same goal: lessening the burden of rendering pixels to let new game technologies shine.
FSR is much easier to integrate into a game.
The biggest reason it isn’t worth comparing FSR and DLSS is that DLSS can only be used on two generations of very expensive Nvidia GPUs while FSR is available on three generations of GPUs from both companies. Arguably more important, however, is AMD’s explicit goal of bringing FSR to console and PC gamers with less expensive rigs. Microsoft has already confirmed that FSR has been included in their developer tools for Xbox — and not just the next-gen Series X or Series S; FSR can even be brought to Xbox One games. Sony, which always plays things close to the chest, has yet to confirm a timeline on their FSR integration. However, Square Enix’s upcoming game Forspoken, a launch exclusive on PS5, will feature FSR in its PC version, so it’s not the wildest leap to assume the game will utilize FSR on PS5 as well — even without official confirmation.
FSR's many advantages begin to fade if you look at how it actually behaves in comparison to DLSS. To start, FSR takes place in the middle of a game’s processing pipeline, after the initial frame has been rendered and key post processing has been completed. Specifically, the process is applied after traditional anti-aliasing and tone mapping have been applied to an image. DLSS, on the other hand, is used to render that initial frame before other effects are applied based on Nvidia’s ever-learning AI model. What this means is that FSR is much more reliant on a game’s initial anti-aliasing pass than Nvidia’s. This makes FSR an additional tool for developers to work with, whereas DLSS is an entirely new philosophy by which games should be rendered. While it has inherent technical disadvantages, it’s also worth noting that FSR is much easier to integrate into a game than DLSS because of where it sits in the pipeline. Forspoken, for example, had FSR successfully running on the game in less than a day.
The initial reviews for FSR back this up. Digital Foundry and Linus Tech Tips each found FSR’s “Ultra Quality” mode to be visually comparable to other modern upsampling methods while offering a noticeable and worthwhile uplift to performance.
The reason I’m being so technical here is that methods like temporal anti-aliasing and upscaling have been a reality in console gaming for a very long time. Checkerboarding was one of many methods used by the PS4 Pro and Xbox One X to achieve a 4K image, with lots of anti-aliasing used in post to clean up the image — all done to hit a performance target of just 30 fps. FSR is the perfect tool to take anti-aliasing to the next level. Even the most casual Digital Foundry fan knows that high quality Temporal Anti Aliasing is one of the biggest reasons console games look as good as they do.
To use a sports metaphor: As AMD made it, FSR right now is a closing pitcher you bring in during the 9th inning after a Quality Start. Meanwhile, Nvidia wants DLSS to be the pitcher that can throw a perfect game. Yes, a perfectly rendered game running at Native 4K will experience slight hits to image quality when FSR is on, but games that already rely on high quality TAA see basically no degradation in quality. By utilizing FSR to lower the rendered resolution, developers suddenly are given a lot more processing headroom to really show what next-gen gaming can look like.
Insomniac Games are already pushing performance boundaries with PS5 titles such as Spider-Man: Remastered, Spider-Man: Miles Morales, and Ratchet and Clank: Rift Apart. All three games feature a “Performance RT” graphics mode that heavily drops resolution to allow the games to run at 60fps with ray tracing enabled. In both Spider-Man games, released at launch with the PS5, the drawbacks of using Performance RT are more noticeable than in the recently released Rift Apart. Details on buildings pop-in at a closer distance, and at night time some areas like Grand Central Terminal often aren’t lit properly until the player gets close to the building. Crowds and other details are also less dense. But, for most players, the experience is so much smoother and cutscenes look so much better that the tradeoffs are well worth it.
Now, imagine a world where Insomniac didn’t need to make those tradeoffs to deliver both ray tracing and 60 fps, only lowering rendered resolution in the game’s “Fidelity” mode and letting FSR do the work of hitting 4K resolution without having to lose all that other detail that makes the game's world more realistic. FSR could be the technology that makes ray traced games running at 60 fps a standard, not a compromise.
FSR could be the technology that makes ray traced games running at 60 fps a standard.
If you’re a budget gamer, something like FSR is going to make a much bigger difference in your experience than someone with the good fortune to afford a RTX 3080 or RX 6800XT. If you’re still playing off your GTX 1060, FSR is a great way to make sure you can keep playing new games with high frame rates. Depending on how many Xbox developers choose to bring the technology to their Xbox One games, FSR could also mean new life for Xbox One games, or even a way to extend the lifespan of those systems to allow more games to offer cross-generation support than would have without it.
Of course, FSR is also an ever evolving product, just like DLSS. DLSS 1.0 resulted in such dramatic hits to image quality that it was worth avoiding until Nvidia drastically improved the algorithm with DLSS 2.0 (which increased the performance gains DLSS provided). FSR is promising now, but that it supports such a wide and diverse array of hardware at launch means that any improvements AMD makes to the algorithm will have a much bigger impact across the gaming sphere than DLSS. While we wait and see what the first FSR console titles look like, the early results on PC make us believe next-gen gaming is going to be even more impressive than we thought.