If you are not the owner of one of the newest and greatest Best graphics cardAnd don’t worry: AMD FidelityFX Super Resolution It can still increase your frame rate and run almost as efficiently as on modern hardware.according to CapFrameX on Twitter, Compared with FP16, running FidelityFX Super Resolution in FP32 “backward compatibility mode” only incurs a performance cost of 7%.
CapFrameX uses SciFiHelmet, RX 6800 XT, 4K resolution and FSR Ultra Quality presets to test these two modes. Although RX 6800 XT does have FP16 and FP32 functions, you can manually change the code of FSR to run FP16 or FP32.
Especially for RX 6800 XT, the performance difference between FP32 and FP16 is only 7%, which is very small, which is good news for users of older graphics cards that do not support faster FP16 code. But please keep in mind that this test is only based on RX 6800 XT and does not necessarily represent other GPU architectures, so you may see different results for other graphics cards.
FP32 is called single precision floating point, while FP16 is half precision floating point. For many years, FP32 has been the standard format for GPU operations, but some operations have not benefited from the increased accuracy and can run faster in FP16 mode-assuming the GPU supports fast FP16 mode. Basically, FP32 allows larger numbers than FP16 and is useful for more complex workflows. However, FP32 requires twice the memory bandwidth and is not necessary in certain workloads.
This is where the “fast” FP16 comes into play. Many modern GPUs, including AMD’s Vega, RDNA, and RDNA2 architectures, can perform FP16 calculations twice as much as FP32 calculations on standard GPU cores. Nvidia’s Turing architecture also supports fast FP16 operations, but interestingly, the Ampere GPU only runs FP16 code at the same rate as FP32 code. Intel’s Gen11 and Xe architectures also support double-speed FP16 operation.
The adoption of FP16 is still relatively new, which is why you don’t see native FP16 features in older graphics hardware. FP16 performance may match FP32 performance at best, but sometimes its running speed is only a small part of FP32-just like most consumer GPUs have relatively limited support for double-precision FP64.
FSR natively supports FP32 and FP16 to ensure that the upgrade and image enhancement technology supports the widest possible range of graphics cards. The performance of the newer architecture in FP16 mode may be better, but compared with the 50% or more frame rate improvement that FSR can provide by upgrading lower resolutions, the 7% performance improvement relative to FP32 is dwarfed.