It's an extremely common misconception that because GPUs have matrix matrix and matrix vector products in the standard, that means GPU companies must be acceler

It's an extremely common misconception that because GPUs have matrix matrix and matrix vector products in the standard, that means GPU companies must be accelerating them. In fact, because it is SIMD across the shader cores already, you can't necessarily do this. Some GPUs do, some don't