For the better part of the last several years, Nvidia's DLSS has been the gold standard for AI-driven upscaling. When it launched alongside the RTX 20-series, it was a rough proof of concept that wasn't rated by anyone, but by DLSS 2.0 it had become a genuine game-changer with way sharper output, better temporal stability, and a meaningful performance boost without the visual trade-offs that traditional upscaling methods demanded. Every generation since has refined that formula further, and DLSS 4.5 represents the most polished version yet.

AMD's FSR has taken a different path. The original FSR was rough around the edges just like DLSS, but it hasn't progressed at the same rate DLSS has. For most of FSR's life, the general consensus has been that it's good, it's open, it's available on more hardware due to its unreliance on specific hardware (until now), but it's not quite DLSS, and Nvidia has always held an image-quality edge over the Red Team. Going into this comparison, I expected that narrative to hold, but I was shocked at just how good FSR has become.

FSR has had some catching up to do

The DLSS suite has only become more impressive

For the better part of the 2020s, FSR and XeSS have played second-fiddle to DLSS in terms of pure upscaling, and that's largely thanks to Nvidia's head start they've had. They were first to introduce the technology, and while it was rather unimpressive to start, it grew into something that Nvidia says 92% of gamers have enabled, some of which certainly doing so out of necessity.

FSR has really only been something users turn on if they don't have an Nvidia card, and it has been measurably worse in terms of image quality until FSR 3, where the gap shrunk the most it ever had. It still wasn't DLSS-level, but it generally saw a slightly larger performance gain in GPU-bound scenarios in return for slightly worse image quality.

I compared FSR 3, which would run on my RTX 5080, to DLSS 4.5, and you can see some clear differences in the foliage in Battlefield 6. The middle of the tree is a lot less defined with FSR 3, and while it still looks decent in motion, during stationary scenes you can absolutely tell the difference.

Head-to-head performance gains are similar

Both upscaling solutions offer substantial performance boosts

I tested both upscalers across a handful of titles at 4K with quality-mode upscaling enabled in a real apples-to-apples scenario, using the same test bench for both, only swapping the GPU between tests, and the numbers are pretty much the same between them. I tested Battlefield 6, Cyberpunk 2077, GTA V Enhanced Edition, and ARC Raiders. The RX 9070 XT and RTX 5080 do perform differently in terms of average FPS, but I wasn't comparing the performance of each card to one another, just to their own performance with upscaling on and off. I also didn't enable frame generation on either card.

This is a real-world use case: booting up the games, turning on FSR or DLSS in quality mode, and playing. They both provided about a 20% uplift over native 4K rendering with TAA in the titles I tested. Neither solution had a consistent edge over the other in raw frame rate. In some titles, DLSS pulled slightly ahead, in others FSR did, and the margins were small enough that they're within the noise of normal benchmark variance. The performance comparison was mostly done out of curiosity, and what I'm really interested in is the image quality.

Image quality is variable between the two

But it's indistinguishable in motion

I was truly shocked at just how close the image quality was between the two upscalers. In motion, you wouldn't be able to tell the difference between them. Standing still in a scene does expose very slight differences in clarity, and these are mostly seen in foliage and trees—entities that upscalers have historically struggled with in the past.

In the GTA V Enhanced Edition example above, you can see that the trees are just the slightest bit more clear at the edges, and in motion this is also the case. Generally, both do a really solid job.

In Cyberpunk 2077, the two upscalers are indistinguishable from one another, with one asterisk. On my setup, the blurriness produced by FSR 4 is much more noticeable during periods of fast motion than the blurriness that's seen with DLSS 4.5. It was really only noticeable while driving quickly through the streets, but it wasn't something I experienced with GTA V.

During online play in Battlefield 6, and ARC Raiders, it was really hard to notice the difference between the DLSS 4.5 and FSR 4. This screenshot isn't an apples-to-apples comparison of the same scene, but I think it shows where each upscaler is particularly strong. The FSR 4 image maintains the foliage quite well when the background is clear of debris, but the quality falls a bit with the smoke in the backdrop. DLSS, by contrast, does a good job despite the smoke behind the tree, but the way the sharpness filter plays with the ground terrain makes it look a bit too sharp for my liking. FSR does a better job maintaining the original quality of the ground terrain in this particular example.

And that's a good microcosm of the general takeaway I had from this exercise: one upscaler isn't necessarily worse than the other in terms of quality consistently across all examples. They provide a slightly different image depending on the game as well, and while you can't use FSR 4 on Nvidia hardware or DLSS 4.5 on AMD hardware, I wouldn't feel left out if I were in either camp exclusively.

In terms of upscalers, the gap has closed

For spatial upscaling, AMD has reached parity with Nvidia in the ways that matter most. Image quality is indistinguishable during gameplay, performance uplift is comparable, and the visual artifacts that used to clearly separate the two are no longer a factor. That doesn't mean the broader technology stack is comparable; DLSS and FSR Redstone are still separated in terms of frame generation artifacts, ecosystem adoption, and ray regeneration. AMD still has catching up to do on those fronts, but at least in terms of pure upscaling, they've caught up nicely.