Supply of next-gen anything gaming is basically impossible to grab, so it was hard to get excited about the announcement of a new graphics card from AMD this week. But what the new graphics card came with was information about FidelityFX Super Resolution, better known as AMD’s alternative to Nvidia’s DLSS black magic.
If you’re out of the loop, DLSS is a neural network-powered technique that’s designed to upscale your games from lower resolutions into higher ones. This gives you the benefit of a higher resolution (sort of) without the same cost to performance.
That’s the theory, although in this case the theory works out pretty damn well in practice. Cyberpunk 2077 is literally unplayable in 4K — especially with ray-tracing — without DLSS, and the game is absolutely something that should be played in 4K. Death Stranding is astonishingly good, with some (but not all) details looking better upscaled to 4K than at their native resolution. And Control‘s latest implementation of DLSS is excellent, even at 4K with all the ray-tracing bells and whistles maxed out.
And with the improvement to FPS, it’s no surprise that people have been waiting for AMD to catch up. After all, the next-gen consoles are built on AMD’s RDNA architecture, so any improvement in PCs might also help boost performance on the consoles, too.
AMD’s formal announcement of the Radeon RX 6700 XT, which you can rewatch above, mentioned FidelityFX Super Resolution but it didn’t offer any details as to when it’d be coming, or why. But Linus Tech Tips got a statement from the company as to what’s taken AMD so long, and it’s fairly logical.
“Rather than rushing it out the door on only one new top-end card, they want it to be cross-platform, so they want it running on all their GPUs — including the ones inside consoles,” Linus said.
Given that AMD’s working with DirectML, the machine learning component within Microsoft’s DirectX 12 framework. Leveraging that opens the door for any advancements to be brought to at least the Xbox Series X, although the architecture of the PS5 means it should theoretically work there too.
DirectML launched two years ago, but we’re yet to really see its wider adoption. And while there was talk that AMD might release FidelityFX via a driver update in the first quarter of the year, it’s pretty clear that the technology is tricky to implement.
One of the biggest problems with DLSS’s early days was adoption. The neural network had to be trained on a game by game basis using a supercomputer, which meant very few games ended up with DLSS support at all. That was later changed to a more generalist approach last year. And with Nvidia publishing a DLSS plugin for Unreal Engine, we should see a lot more games incorporate the AI-powered super sampling in the months and years ahead.
AMD has always preferred solutions that can be enabled very simply, so the delays here make sense. If AMD can remove as many barriers as possible to its implementation, gamers will ultimately benefit. There’s little value in a technique that only one or two games use, as impressive as they might be. We’ve already seen how effective Nvidia’s neural network can be — and once AMD gets their alternative into gear, the whole industry will benefit.