FSR 2.0 tested: AMD’s new upscaler closes the gap on DLSS

AMD’s FSR 2.0 upscaler is out now – in just one game, Deathloop, but still. This is a big moment for the heavily upgraded FSR (FidelityFX Super Resolution), as it basically tries to combat Nvidia’s rival DLSS (Deep Learning Super Sampling) with some of its own tactics. There are no improvements to Nvidia’s machine learning, but while FSR 1.0 was a simple spatial upscaler — meaning data used to scale up a frame can only come from the frame itself — FSR 2.0 uses a temporal upscaling technique similar to that of DLSS. This also allows it to extract data from earlier frames, producing a sharper, higher quality image.

The lack of machine learning also has the advantage that FSR 2.0, like FSR 1.0, remains compatible with each fairly recent gaming GPU. Nvidia might make some of the best graphics cards of this generation, but you don’t need something brand new and super powerful to let FSR 2.0 boost your frame rates. To find out how well it works, compared to both DLSS and FSR 1.0, I reinstalled Deathloop and gave all three upscalers a spin.

FSR 2.0 Tested: Deathloop 4K Performance

Upscaling works by displaying games at a lower resolution than your monitor’s native resolution, then reconstructing each frame to make it look as close as possible to its native resolution—sometimes even better. This helps take the sting out of demanding resolutions like 4K, while it’s not so good for 1080p because the display resolution will be so low that even a good upscaler will struggle to make up for the sharpness deficit.

I tested Deathloop’s upscalers on an RTX 3070 GPU, paired with an Intel Core i5-1600K and 16 GB of RAM. Before we get into FPS gubbins, though, let’s take a look at how FSR 2.0 improves on 1.0’s overall visual quality.

FYI, “Quality” is the highest possible setting for both FSR 2.0 and DLSS; with FSR 1.0 “Quality” is in fact the second highest, with “Ultra Quality” the best. “Balanced” is the second highest for FSR 2.0 and DLSS, making it the equivalent of “Quality” on FSR 1.0. Not confusing at all.

Left: FSR 2.0 on quality. Right: FSR 1.0 on Ultra Quality

Other than that, FSR 2.0 is definitely an aesthetic upgrade from FSR 1.0. Fine details that would previously be blurred by the scaling process are sharper, while straight lines — especially on distant objects — don’t exhibit as much erratic aliasing as on 1.0. It’s worth noting that FSR 2.0 is again scaling up the anti-aliasing setting that the game has on at the time; it doesn’t include its own AA, like DLSS does. In the case of Deathloop, I used the Ultra preset for testing, so in any FSR or native shots you’re looking at TAA.

A comparison image with a snowy scene in Deathloop.  On the left, the scene is rendered with FSR 2.0, on the right, it is rendered with DLSS.

Left: FSR 2.0 on quality. Right. DLSS on quality

An improvement over FSR 1.0 means FSR 2.0 is much closer to DLSS in fidelity… just not enough to top it. Frankly, the difference isn’t often as noticeable as when we zoom in this close, but you can tell from distant details (like that oil barrel and the thin metal grilles on the generator) that Nvidia’s technology remains slightly superior. Even if you specifically need a GeForce RTX card to use it.

In motion, I also noticed less shimmer on objects like stairs and vehicle tracks than I did on FSR 2.0, and that red puff speaks for itself. It’s oddly pixelated with FSR 2.0, while DLSS keeps a more natural look.

A comparison image with a snowy scene in Deathloop.  On the left, the scene is rendered in FSR 2.0, on the right, it is rendered in native 4K.

Left: FSR 2.0 on quality. Right: native 4K

Still, at least FSR 2.0 can join DLSS in the Better Than Native club. Compared to rendering straight to 4K, FSR 2.0 seems to improve sharpness and even texture details. In the comparison above, you can see how the leather and stitching on Colt’s glove, as well as the rotting brick wall, look just a bit sharper with FSR 2.0 on the quality setting.

A bar graph showing how well Deathloop works in 4K, with different scaling settings.

At 4K, Deathloop’s performance gains from upscaling are unusually modest. The FSR 1.0 quality setting is the fastest of the bunch, although using the higher quality FSR 2.0 or DLSS options only adds about 10% more FPS than native 4K. On the plus side, AMD card owners will be relieved that the newer version of FSR isn’t significantly slower than DLSS, and there’s also some reassurance in knowing that you can select the quality option and still get nearly identical performance to the Balanced option. It’s even with the equivalent Ultra Quality setting of FSR 1.0, so the visual upgrade doesn’t necessarily come with a performance cost.

FSR 2.0 tested: Deathloop 1440p performance

Until now, you could never call FSR 2.0 a DLSS killer, but it’s quite a step forward at 4K. On the more widely used 1440p, I’d say it’s even better, especially when compared to its predecessor.

Check out the what-the-hell-vehicle-this-is below; almost everything from the aliasing around the door edges to the detailing on the tracks and even the matte texture of the bodywork looks better with FSR 2.0. The only reason it looks a bit hazy is because of a passing gust of snowy sky.

A comparison image with a snowy scene in Deathloop.  On the left the scene is represented with FSR 2.0, on the right it is represented with FSR 1.0.

Left: FSR 2.0 on quality. Right: FSR 1.0 on Ultra Quality

You can also see how the brick wall, which is closer to the camera, has much more defined edges.

A comparison image with a snowy scene in Deathloop.  On the left, the scene is rendered with FSR 2.0, on the right, it is rendered with DLSS.

Left: FSR 2.0 on quality. Right. DLSS on quality

FSR 2.0 still has a pixelated smoke issue at 1440p, and again DLSS Deathloop’s quality setting improves a bit more overall. To its credit, FSR 2.0 can look even sharper than quality grade DLSS in certain places – just look at the graffiti on the yellow truck – but it’s a tougher, more artificial sharpness.

A comparison image with a snowy scene in Deathloop.  On the left, the scene is rendered using FSR 2.0, on the right, it is rendered in native 1440p.

Left: FSR 2.0 on quality. Right: native 1440p

Oh yeah. FSR 2.0 still manages to look even better than native at 1440p – or at least sharper. Once again it performs its texture-polishing trick on Colt’s gloves, while further details are given a touch of extra sharpness. Then of course there’s the performance bonus…

A bar graph showing how well Deathloop works at 1440p, with different scaling settings.

Interestingly, the old FSR 1.0 is the worst performing Deathloop upscaler at 1440p, despite advancing at 4K. What’s more, FSR 2.0 takes its place and is ahead of DLSS for both their quality and balanced settings.

Considering that DLSS still looks a bit nicer, I’d say it’s still the way to go if you already own a GeForce RTX GPU. Which, without disrespect to AMD, shouldn’t be a very surprising outcome: RTX cards have machine learning hardware aimed at making DLSS the ideal upscaler.

However, that’s not to say that FSR 2.0 is a failure. Even if it can’t take advantage of dedicated hardware, it has managed to significantly close the quality gap on DLSS, while maintaining cross-platform quality and beating even the simpler, less intensive FSR 1.0 with 1440p performance. It’s a much improved bit of resolution trick, and if DLSS isn’t available I’ll find it in a game’s graphics settings instead. Once more than one game starts supporting it anyway.

Leave a Comment