definitely! it has the advantage that it can run CUDA kernels but on the other hand it has lower memory bandwidth and probably loses a token/s fight for many LLMs.