Patent Suggests Future AMD GPUs Could Support Variable Rate Shading


Staff member
Mar 3, 2018
Twitter user 0x22h spotted a patent suggesting that AMD could introduce Variable Rate Shading in their future GPUs. Nvidia's Turing architecture already supports VRS, which AMD's patent succinctly describes as "a technique for performing rasterization and pixel shading with decoupled resolution..." In other words, different parts of a scene could natively be rendered at different resolutions. Such a technique would be particularly useful with high resolution, high density displays, where rendering the whole scene at 4K might not be necessary, and it could give future consoles a higher-quality alternative to rendering games at non-native resolutions. Theoretically, it could also provide an easier way for developers to support foveated rendering in traditional displays and VR headsets that support eye tracking, and PCGamesN mentions that Nvidia already intends to use VRS for motion adaptive shading.

Nvidia's Turing GPUs, found in both the high-end RTX 2080 Ti and the lower-caste GTX 1660 Ti, uses two key algorithms to take advantage of VRS - motion adaptive shading and content adaptive shading. The use of motion adaptive shading is most obvious when applied to something like a racing game, or something like Grand Theft Auto, where there is a huge amount of motion. Things in the middle of the screen, such as the car or character, which aren’t be subject to a lot of relative motion will be rendered in full. Everything else that is moving rapidly can then be rendered at a lower rate as they don't need the same level of detail. Our eyes can't perceive the pixel quality difference when they're flying past at speed anyway.
I am not sure how stuff like this will play out in the real world when implemented but the fact they are working on this sort of stuff is really cool.
How does the VRS work with the RTX 2080Ti and GTX 1660 Ti is retroactively enabled by default is it toggled on a case by case basis depending on the game? It could easily be viewed as a cheat if it's retroactively on by default for example and impacts visual quality. I actually think you could do this with post process injection and AA or other post process effects with ReShade manipulating the split screen effect. For example the following type of thing could be adapted I believe tweaking the splits a bit further and just applying a bit more AA to a customized focal area which could be adjusted to taste even.

The adaptive shading/post process has great applications if could enhance things selectively in a nice way. For example use different split points to apply different effects or different strengths to effects to different split area's for gradient post process shading. The nice benefit is you don't need to wasted as many resources on a effect covering the entire screen, but rather custom regions of it which is a nice way to scale things back to conserve resources.