AMD FSR 2.0 Source Code Hits GPUOpen With Samples And A Demo, Here’s How To Try It
by
Zak Killian
—
Wednesday, June 22, 2022, 04:58 PM EDT
NVIDIA's DLSS is one of those few technologies that comes along every so often and legitimately delivers the seemingly impossible: more with less. AMD's first attempt at challenging DLSS wasn't quite as advanced; while the original FidelityFX Super Resolution had its advantages, the image quality wasn't always in the same ballpark.
FSR 2.0 is a whole different story, though. While it still doesn't exactly match DLSS in every instance, it looks fantastic, and it runs on any graphics processor capable of DirectX 12 or Vulkan. AMD promised that it would take "five minutes" to implement if your game already supports DLSS, and indeed several titles have picked it up already.
The warehouse scene is much less demanding than the Sponza palace scene.
That said, if you're not playing God of War, Deathloop, or Farming Simulator 22, you probably may not have seen FSR 2.0 in action. If you'd like to do so for yourself, head over to the GPUOpen project and download the sample package. It's the button labeled "FSR 2 sample." This 917MB package includes sample applications for both DirectX 12 and Vulkan.
Using the sample apps, you can fly around the familiar Sponza palace scene or a simpler abandoned warehouse scene, while fiddling with the lighting as well as the tone mapping and various forms of upscaling. DLSS isn't supported, unfortunately, but you can compare FSR 2.0 against native resolution rendering, FSR 1.0, or simpler upscalers using bicubic sharpening, bilinear filtering, or straight-up point sampling.
Deathloop detail comparison. Clockwise from left: Native, FSR1, DLSS, FSR2.
This sample package is merely a side effect of the real news today, which is that AMD has released the source code for FSR 2.0. That means that developers working on DirectX 12 or Vulkan applications can independently integrate the smart upscaler into their own software. Doing so is nontrivial, but as mentioned above, the same work applies to implementing DLSS, so it's ultimately a win for everyone. You can find the source on Github.
