@hisham_hm @justincormack @lproven @kmett ... while also organizing the flow of data (GBs of data) such that you exploit locality in a way that you don't get slowed down by the memory hierarchy. And then there are vectorized operations, and nowadays GPUs of course, and so on.