AMD Navi 31 GPU FP32 delivers 92 TFLOPS of performance
AMD has major changes in the next-generation architecture specifications, but what exactly is still uncertain, assuming AMD retains the 60 WorkGroup Processor (WGP) and 15360 stream processor configuration, a core speed of 2995MHz is required to achieve 92 TFLOPS of FP32 single-precision computing performance. However, similar to NVIDIA’s Ada architecture, AMD may introduce a different 32bit instruction matrix to improve floating-point performance.
Also contrary to previous rumors, AMD may still keep Compute Units for its RDNA3 (GFX11) architecture. Twitter user @Kepler_L2: found that the AMD GPU driver change explained this: