AMD's Radeon RX 7000 GPUs could feature hardware AI processors similar to Nvidia's Tensor cores, as suggested by latest Linux patches
Twitter user Kepler_L2 recently spotted the addition of AI-ML instruction support for the upcoming RDNA3 Radeon RX 7000 GPUs from AMD. The addition was revealed in the latest Linux patches for the GFX11 architecture, which is AMD’s codename for RDNA3. More specifically, the patches include Wave Matrix Multiply-Accumulate (WMMA) instructions that are used to operate on large sets of numbers, especially in machine learning workloads. Is AMD finally adding AI-enabled hardware in their Radeon GPUs?
As pointed out by Videocardz, the AMD’s CDNA architecture already supports WMMA instructions, but CDNA only powers compute GPUs like the Instinct MI200. The inclusion of WMMA in gaming GPUs may suggest that AMD has been working on an alternative to Nvidia’s Tensor cores that are mainly used to process the DLSS image supersampling algorithms.
AMD’s FSR 2.0 already proved that image supersampling does not necessarily need AI to deliver good results, yet the inclusion of AI-powered hardware in the upcoming RDNA3 GPUs could bring major changes to future FSR versions. Nevertheless, AMD should consider keeping this standard open source, or even better, make it compatible with Nvidia’s Tensor cores in order to avoid complicating things for game developers.