The workload is . It thrives on wavefront size (64 threads on AMD RDNA architectures, 32 on NVIDIA). The library is optimized to maximize cache hits on the history buffer lookups.
While ffx_fsr2_api_x64 does the heavy lifting, it is a "Garbage In, Garbage Out" system. The API demands specific inputs from the game engine, and if those inputs are flawed, the result is a shimmering, ghosting mess. ffx_fsr2_api_x64
When a game engine (like Unreal Engine, Unity, or a proprietary engine) wants to use FSR, it doesn't execute the upscaling logic directly in the game's main loop. Instead, it links against this API. The library acts as a "middle manager" that handles: The workload is