2 ms·

I’m rather hoping that the cpu cache hierarchy will one day be as programmable is the GPU local, shared, global memory.