As part of the Games Developers Conference (GDC), Microsoft has announced an extension of DirectX 12. Thanks to the technology called "Variable Rate Shading" (VRS), it is necessary to save on a GPU a processing time such that the second games Microsoft has a higher frame rate of 14-20 percent. As a proprietary development called "Coarse Pixel Shading" (CPS), Intel had already demonstrated the trick in December 2018 for its next integrated GPU of the 11th generation and spoke at that time up to 40 percent of fixed graphics.
With VRS, which is standardized in DirectX 12, the technology should now be available for all graphics processors. Nvidia already supports the Turing graphics card process, it is used between others in Wolfenstein 2. Developers must customize their games specifically for Variable Rate Shading. Microsoft speaks in his blog of "a few days", which are necessary for this. If and as the functionally supported VRS graphics hardware, Microsoft has not yet announced, just as a corresponding announcement from AMD is still pending. However, Radeon's producers have recently released their VRS patents.
Not exactly, but much faster
Behind all these approaches is the idea of not calculating every pixel with the utmost accuracy of all shaders. VRS now combines several pixels into blocks that are calculated together. This reduces the calculation accuracy in the areas treated with VRS in the image, but not the actual screen resolution, and can save calculation time. VRS has three methods to do this: either each draw call can use VRS, or within a draw call only part of the image can be handled, or within a draw call only some triangles are processed, aliasing between the exact and inaccurate areas should prevent. All three methods can be combined.
There are two fundamentally different approaches that Microsoft calls "Tier 1" and "Tier 2". The first method deals with different drawing calls with different pixel shading resolution. For example, developers call to design a blurred background that can be handled more inaccurately during a call and the next call calculates the foreground in more detail. Apparently, we mean scenes like wide landscapes, in which the player concentrates on his own figure and on nearby objects.
Level 2 requires a precise pre-construction of the image, the so-called "screen image". It is also used by many other modern methods such as Ambient Occlusion Shading. This allows VRS to also determine dynamically calculated areas with less accuracy. Microsoft calls as a first-person shooter application, in which the center of the screen attracts the attention of the player.
Intel's demonstrations, also cited by Microsoft, demonstrated this with a three-level accuracy based on the technological demonstration of the unreal "Sun Temple" engine. Level 2 is slower than level 1, but the quality difference is almost imperceptible. A customized version of Civilization VI should have increased by 20% with level 1, while level 2 was still 14%. The fact that VRS games suddenly look worse with a patch is, according to the current situation, hardly to be feared. So, Intel said about the presentation of its implementation that the function in the driver should be disabled by default.
Game developers should be able to get started with VRS as soon as possible. Microsoft plans to release the specifications, code samples and slides of the presentation in its DirectX forums after the GDC presentations. Incidentally, VRS should not be confused with VSR, the latter called "Virtual Super Resolution" the AMD method for calculating subsampling games with a resolution higher than that supported by the monitor. This serves as an alternative or extension to anti-aliasing features.