4 ms·

Ignoring the cache hierarchy on a CPU for matrix multiplication gets you a 100x performance drop, just like a GPU