*Rocky theme* It's the, eye of the upscaler, it's the thrill of the... performance comparison?
When it comes to upscalers, there's little doubt that Nvidia's DLSS has become the go-to choice when compared to its nearest rivals—AMD's FSR and Intel's XeSS. Image quality comparisons have repeatedly shown that Nvidia's upscaling solution simply preserves detail better than the others, with less flickering, shimmering and graining—and often better performance to boot.
However, AMD has recently released FSR 3.1, a supposedly much improved version of its own upscaling tech. Unlike DLSS (and like previous versions of FSR) this latest release is still graphics card agnostic thanks to its compute-based approach to upscaling.
While FSR 3 added its own version of Frame Generation, the fundamental upscaling tech remained the same as 2.2, putting it on the backfoot compared to Nvidia. However, the release notes for FSR 3.1 promise much better temporal stability, less ghosting, and improved upscaling quality overall. Bold claims—and worth a spot of testing, we thought.
So, time for a head to head then. Can FSR's latest release bring it near to—or even surpass—DLSS and its excellent upscaling chops, and what does image quality look like in comparison? And which, crucially, is best for performance?
I've equipped my machine with an RTX 4070 to find out. Let's kick things off with Horizon Forbidden West. A graphically beautiful game, for sure, and one that's filled with lots of small flying insects, running water, and flowing hair. All things that traditionally give upscalers trouble. In fact, Horizon FW gives me a bit of a gift here: the very first opening section of the game puts you in a small area filled with all of the above.
Very kind of it, I think you'll agree.
As such, I placed Aloy on an appropriate log to capture some of the fast moving beasties, before running a few laps to see how both upscalers handle some motion. The above footage shows the game running at maximum settings at 1440p with FSR 3.1 and DLSS 3.5 set to Quality, with motion blur off.
YouTube compression is a factor here, so I suggest whacking the quality up to full for the best results. At first glance, both FSR and DLSS do an admirable job of preserving image quality, and while Aloy is surrounded by small flying insects, the unwanted trailing effect of previous FSR versions does appear to be much improved.
Pay close attention, however, to the side of Aloy's face when I rotate the camera in the flower patch at around 0:41. While both upscalers show a little blurring, FSR 3.1 is still noticeably worse, with a longer trail of blur under slow, consistent movement.
Something I'm particularly used to seeing with FSR is a noticeably grainy or "fizzy" effect surrounding objects in motion compared to DLSS. While this also appears to be improved, it's still noticeable at 1.02 when Aloy jumps in front of the water. Silhouetted against the fast moving stream, there's noticeable distortion and graining around her frame under movement, an effect that's much less pronounced with DLSS enabled.
Still, YouTube compression being what it is, perhaps we should take a look at some still frames for a dive into the details. Below are screenshots taken using FSR 3.1 and DLSS 3.5, also at 1440p quality settings. Both upscalers do a pretty good job here, although I managed to capture part of Aloy's idle animation where her head was turning to get two comparable motion screenshots (no easy feat, let me tell you).
While FSR 3.1 performs admirably, and I appreciate it can be difficult to see on a small screenshot, there is noticeably less blurring on the side of her head in the DLSS version. DLSS also looks to do a slightly better job on some of the static background details, although it's much of a muchness for the most part. At 1440p at least, FSR 3.1 does a comparable job, although there are tell-tale signs that it's still struggling with details and graining compared to Nvidia's solution.
Drop things down to 1080p performance settings on both upscalers and zoom in a bit, however, and well...I'll let you decide.
Eesh. While I appreciate Aloy's head doesn't line up perfectly from shot to shot (you really don't know the fun I had putting these together), it's pretty clear where FSR 3.1 is failing here, and DLSS is succeeding. Motion is a difficult thing to capture in a screenshot, but believe me when I say that Aloy's head is moving in both frames, and that DLSS does such a better job at performance settings, it's not even close.
When it comes to performance itself, FSR 3.1 actually puts on a decent showing—compared to DLSS 3.5, at least. While the figures here are within margins of error, I performed multiple runs to check my results and found that, on average, FSR 3.1 usually manages to eek a couple of frames ahead.
That's a pretty impressive result—although as my image comparisons show above, some visual sacrifices have been made. It's nice to know that, if performance is your concern, AMD's upscaler can deliver comparable, even slightly better results than Nvidia's—at least in this game.
There's always Frame Generation to consider too, both Nvidia and AMD flavors, although I kept it off for all the comparisons here for the sake of a raw performance comparison.
Which means we'd better move on. Ratchet & Clank: Rift Apart also features both FSR 3.1 and DLSS, although this time it's the latest revision, DLSS 3.7. It's got some fluffy, fuzzy characters amidst jam-packed set design, and even some ray tracing shenanigans to give both upscalers a real workout.
So, I whacked the settings up to maximum at 1440p, including ray tracing, and set to work Ratcheting and Clanking—or whatever it is you do in this game.
Again, FSR does well at these settings. Still, that graining effect is perceptible under motion, and points where I turn the character to face the camera show that Ratchet's furry ears are susceptible to a pixelated effect against the background that's much improved with DLSS. Nvidia's solution seems to be better at accurately smoothing crunchy edges with detailed scenes behind them, and switching back and forth between the two reveals a smoother, more pleasant-looking image in the DLSS version.
Ghosting, however, does appear to be less of an issue for FSR than it used to be. There's lots of floating ticker tape, fast moving whizz-bang effects, and smashy, crashy particles in this scene, and I didn't notice too much of a trailing or ghosting effect. Still, for image quality overall? DLSS produces smoother-looking frames, with less artifacting.
It's the same story again: both are perfectly usable, both produce an image quality that, if you're not paying too much attention to the details, would absolutely deliver a good experience for your average player. That being said, it's still obvious that DLSS is producing better images in detail-packed scenes at 1440p, both under motion and standing still.
For a side-by-side comparison though, let's drop things down to 1080p performance mode and collect some screenshots, have a bit of a zoom, and see what's going on.
Look what FSR did to my boy. Ratchet looks, quite frankly, unwell in the FSR version. A lot of clarity is lost in the eyes, while the arms show significantly more artifacting in the FSR 3.1 version compared to DLSS. I'd also take a close look at the shrubbery behind Ratchet's ear—in the DLSS shot, plenty of fine detail is preserved, whereas FSR looks almost cell-shaded by comparison.
It's muddy, pixelated, and simply looks like a different set of textures entirely.
Fast motion at these settings was simply unpleasant to watch. In fact, things were so bad using FSR in performance mode at this resolution that I recorded an extra video showing the effect:
Yep, even video compression can't hide the flaws of that one. Fizzy, grainy, and so distracting that I felt myself grimacing as I played. If your system needs 1080p performance settings to get a good frame rate out of this game, I simply would rather not play it in this state than use FSR performance mode. DLSS doesn't look wonderful either at these settings, but the image is much smoother, and simply less distracting to look at.
And then we come to performance. At 1440p quality, FSR on average produced an identical result. The 1440p balanced and 1080p performance figures, however, reveal a different story. Here, DLSS pulls comfortably ahead, all while delivering an image quality that's simply much nicer to look at.
So, what have we learned? Well, while FSR 3.1 undoubtedly handles motion better than its predecessors, and delivers decent image quality at 1440p quality settings, compared to DLSS it's still the worse of the two. No getting around it, I'm afraid—if you have an Nvidia card, there's nothing to recommend using FSR over DLSS, other than perhaps a few frames in certain games.
For those of you on AMD or Intel hardware though, you will likely see an improvement over previous FSR versions in games that upgrade to the latest release. I've been experimenting with various iterations of AMD's upscaler for a while now, and I can tell that improvements have been made.
...there's still that nagging, teasing doubt that, if I was playing on an Nvidia card, my games would simply look a lot nicer using DLSS
But ultimately, AMD's compute-based, hardware agnostic solution still loses out to Nvidia's highly-refined, AI-based iteration.
Perhaps that should come as little surprise, but part of me was hoping—like many I'm sure—that this update would finally solve some of the graining, fizzing, and for want of a better term "FSR-like" artifacting that has plagued every version I've tested to date. And it simply hasn't. Maybe I'm expecting too much from AMD's solution given the difference in methodology here, but as with GPU performance overall, Nvidia still has AMD firmly on the backfoot.
As an AMD-card user, I'll take some comfort in the fact that games using FSR 3.1 will indeed look a bit better with AMD's solution enabled than previous. But there's still that nagging, teasing doubt that, if I was playing on an Nvidia card, my games would simply look a lot nicer using DLSS. And in some, I'd be getting significantly better performance, too.
Thems the facts. This round is over, and DLSS stands victorious. AMD might be improving, but even after putting up a brave fight with this latest FSR iteration, Nvidia remains the upscaling champ.
