if it matches the CUDA performance on FP16, NVProfile report Tensor Cores not being used, AND the Radeon Vega 64 is close to the RTX 2080 S ( Video Enhance AI v1.7.0 - #370 by bnumerick ), then no Tensor Core is being used (Vega 64 is closer to 1070 performance at AI - ie Tensor Instructions -, while roughtly matching RTX 2080 S at FP32 and FP16). Also if it was using Tensor Cores, Intel iGPUs would get a big uplift aswell (ever since Haswell intel iGPUs have Tensor instructions - so it literally works as a Tensor Core)
