Another year, another supersampling battle! Though the names of the companies are the same, this time around in 2022 it’s AMD stepping up to the plate with its answer to Nvidia’s Deep Learning Supersampling (DLSS) frame-rate-boosting technology. Team Red’s new entry is called AMD Radeon Super Resolution (RSR). This won’t be the first time we’ve pitted Nvidia’s and AMD’s various sharpening and supersampling features against one another (nor will it likely be the last). But it is the first time, we suspect, that AMD has a real chance of undoing the outright dominance DLSS holds in this esoteric graphics arena.
As a quick refresher, here is what supersamplers do. In essence, they allow you to run a game on your PC at a lower resolution than the number of pixels in your monitor, then “supersample” the image back up to a quality that should seem no different from the native resolution to the naked eye. The idea here is “adding performance for free,” though the reality of how it all works is way more complicated than that. Let’s get into it.
AMD RSR: Supersampling for Most Modern Games
AMD’s first attempt in supersampling, dubbed FidelityFX Super Resolution (FSR), works well enough already, to be sure. Support for FSR, however, needs to be integrated on a per-game basis, just like Nvidia’s DLSS. And that has proved to be the weak point of both so far.
Because it’s up to a game developer to get FSR or DLSS working, adoption of both has been slow. FSR, for example, has been on the market for six months, and it’s been added to just 80 titles in total. (DLSS is in more games, but it is far from ubiquitous.) On the flip side, though, FSR holds a big advantage: It works with almost every graphics chip launched in the past several years, from AMD and Nvidia, thanks to its shader-level implementation.
Here’s a quick refresher on where things stand in the world of supersamplers in 2022…
Conversely, AMD says its driver-level RSR service will supersample almost any game that runs on either DirectX 11 or DirectX 12/12+, using the same algorithm that powers FSR 1.0. For those keeping count, this immediately makes RSR the most widely compatible supersampler, on a game-by-game basis, by a factor of…well, a lot. Think of every game that’s used one of those two APIs since DX11 went stable in 2009, and you start to get a sense of the scope that AMD is shooting for here.
The catch (and there’s always a catch when it comes to supersampling tools like these, it seems) is that RSR, unlike FSR, will work only on AMD GPUs, and more specifically, only AMD GPUs from the Radeon RX 5000 Series and later. Much like DLSS when it first launched on Nvidia’s GeForce RTX-badged cards, AMD has limited its most valuable supersampling tech to a subset of hardware, one that excludes even that segment of gamers still using Polaris- and Vega-era AMD graphics cards.
The move to integrate tens of thousands of titles into the ecosystem on release easily surpasses Nvidia’s current count of around 150 DLSS-compatible titles (as of March 2022). That should, if the quality testing holds up, make RSR the de facto choice for any owners of recent AMD GPUs to try moving forward.
So, how does RSR hold up under scrutiny? Is this truly the supersampler for all, or is it just a glorified sharpener that, once you’re close enough to see how the sleight of hand works, gives away the game?
Performance Testing
We decided to test the performance of all three supersamplers using everyone’s favorite game to hate from the past few years, Cyberpunk 2077. Regardless of the problems the game had at launch, these days it’s actually quite stable, performance-wise. It’s also the only title we had on hand compatible with all three supersampling technologies (FSR, RSR, and DLSS) in one. And it’s one of just a handful of titles that’s actually managed to adopt and integrate the latest version of DLSS, 2.3, into its engine.
Finding a game that supports all of the supersamplers in one may be tough, but thankfully, it’s not as hard to find a game that RSR will work with natively, since just about any game that supports DLSS relies on DirectX 11 or DirectX 12/12+. Only a select few, though, have compatibility for both DLSS and FSR baked into the same pie. We’re talking just two games out there, Cyberpunk being one. (The other is Call of Duty: Vanguard.)
As for hardware, we decided to set up a duel of Nvidia’s GeForce RTX 3070 Ti Founders Edition against AMD’s Radeon RX 6800. These two cards are near-equals in power that AMD is aiming to make equal in added features as 2022 rolls on. (By that, we mean RSR is an added supersampling feature to AMD GPUs, like DLSS is an added feature to Nvidia GPUs.)
Also, don’t put too much stock into the numbers here being a perfect 1:1 comparison. That is simply not possible with RSR as a Radeon-only technology; we had to test across two different GPUs to get DLSS in the mix. RSR is compatible only with AMD Radeon RX 5000 or later GPUs; Nvidia’s DLSS is compatible only with GeForce RTX ones.
In defense of our testing here, though, the Radeon RX 6800 and RTX 3070 Ti we used were close in results when we tested them across more than a dozen games at their respective launches. That fact reinforced here by their lock-step finishes of 18fps when we tested the game at our native 4K resolution with all settings (apart from ray-tracing) tuned to Cyberpunk 2077’s Ultra preset via the in-game graphics menu.
So, here’s what we did. We tested each supersampler through three of its most closely aligned resolution presets. While RSR always needs to be set manually—you choose the down-rendered resolution first and RSR handles the supersampling in the background—DLSS controls your resolution for you, the same way that FSR does.
Again, more reasons to grab your shaker of Morton’s salt before we dive into the results…
In every test we ran, even accounting for the minimal hardware power differences, AMD’s RSR continuously proved itself to be the fastest option by a strong margin. But of course, it’s easy to get any game running quickly, as long as you’re willing to sacrifice visual quality…
Quality Testing
The real challenge for all these supersampling solutions is how well they handle the second part of the illusion: adding performance without visually degrading the onscreen image.
To test this, we chose a Cyberpunk 2077 scene that provided some curveballs for the competing algorithms to work through. You can see it below. The tricky bits include hexagons across the back of the vending machine and the cylindrical soda cans, as well as a TV screen in-game that intentionally blurs a part of its own image (the COMBATCAB panel you see).
If the FSR or DLSS algorithms tried to correct that blur, we’d be able to tell that the algorithms were working overtime. Meanwhile, if the lines between the background of the vending machine started blending with the objects in front, this would be another indication of some of the “smudging” problems all three technologies have had to overcome in their various iterations.
Because of the way the various supersamplers work—namely, where they actually apply their effects in the GPU-to-monitor conversation—we had to pipe our video signal through a separate capture card to capture a 4K image, which we then cropped into a data graphing tool for comparison below. We mention this because though there has been some degradation to the image quality presented on page here, all our recordings were processed and evaluated using a full 4K lossless uncompressed format. (In other words, what you see is relative, not an absolute, result of what the supersamplers can do; the images have gone through a data presentation tool, and are also limited by what your own display can show.)
To view the full-screen image in its compressed form below, click the icon that appears in the top-right corner when you hover over the screenshot…
In this testing, we didn’t see anything that outright surprised or shocked us. Whether in a fast- or slow-moving scene, Nvidia’s DLSS remained the best-looking option by a big margin.
Next down the list was FSR, which was close to DLSS in quality but still had many of the same problems we saw when we first tested it back in June of last year. These included visual artifacts around character and object outlines, as well as some of the smudging and haloing problems that were endemic in DLSS 1.0 (but later corrected in 2.0).
RSR, which is based on the same algorithm as FSR 1.0, also exhibited these problems. They were equally as pronounced in low motion, but we noticed smudging even more than in FSR when the onscreen action kicked into higher gear. This is something AMD told us to expect, but it remained a jarring experience once we actually laid eyes on it.
In looking at still images, though, the differences among all three supersamplers are harder to discern. It seems much of the tweaking, refining, and iterating that’s gone into efforts like DLSS 2.0 (and subsequent versions to 2.3) have slowly but steadily gotten the core aliasing elements under control to look great in still images. This applies less so to FSR 1.0 (and by extension, RSR), which still retains some visible artifacts around the edges of characters, and in smaller lines of text. Overall, the effect is more noticeable than with DLSS. But keep in mind that RSR was regularly the winner in kicking up frame rates when it was applied, versus native resolution testing.
Though RSR, to our eyes, may lose this particular subjective-quality battle, we’ve also gathered that AMD’s next rollout of the underlying FSR algorithm for RSR, FSR 2.0, is just around the corner. Once RSR is backed by FSR 2.0, Nvidia may need to come up with something quick to bridge the several-thousand-game moat that AMD will have put between itself and Big Green’s front lawn with little more than a software update. Radeon owners could be happy campers if FSR 2.0 ups the quality level significantly, and RSR benefits.
That’s the future, though. If you just need a supersampler that, as of today, works on tens of thousands of games? RSR, backed by FSR 1.0, isn’t a bad start. Not bad at all…assuming you have a late-model AMD GPU.
Verdict: Nvidia Should Watch Its Rear-View Mirror
Long ago, a rumor noted that Nvidia was experimenting with a broad implementation of what (at the time) was only referred to as “DLSS 3.0,” a version of its supersampling technology that would work on thousands of games already compatible with temporal anti-aliasing (TAA). But in the time since that supposed leak (ostensibly, from a board meeting with Nvidia CEO Jensen Huang) first hit the Net, we’re only up to DLSS 2.3, and AMD just released a supersampler that comes close to delivering on that rumor.
In a time when available graphics cards are scarce and pricey on the open market, gamers are looking for more ways than ever to get the most out of the GPUs they already have. Or, they want to make sure, if they shell out for a new one, that they upgrade to a card that’s future-proofed with all the features they want. Supersamplers, and especially supersamplers like AMD’s RSR that work on almost any modern PC game, help on both fronts.
What started as an exclusive feature for the GeForce RTX-owning elite is now, thanks to AMD’s FSR and RSR, closer to becoming a feature for all gamers. (Then there’s Intel’s announced XeSS and its own coming GPUs; hopefully XeSS will be our last new stop on the supersampler train for a while!) So whether you’re an AMD Radeon-faithful gamer, you pray at the altar of Nvidia’s GeForce, or you just want your games to run faster no matter who makes it happen, there’s a supersampler out there for you.
In the early looking, RSR may not always produce the best-looking image; that privilege still comes at a GeForce RTX-level premium. However, as the algorithms improve—including AMD’s announced incoming upgrade to FSR 2.0, which reportedly brings significant improvements to image quality—we expect Nvidia only has a matter of months before its supersampling dominance could see a real challenger.
