Why is there huge performance hit in 2048x2048 versus 2047x2047 array multiplication?

Question

I am making some matrix multiplication benchmarking, as previously mentioned in Why is MATLAB so fast in matrix multiplication?

Now I\'ve got another issue, when mu

Answer 1

I suspect it is the result of something called "Sequential Flooding". What this is is that you are trying to loop through the list of objects that is slightly larger than the cache size, thus every single request to a the list (array) must be done from the ram, and you will not get a single cache hit.

In your case, you are looping through your arrays 2048 indexes 2048 times, but you only have space for 2047 (possibly due to some overhead from the array structure), so each time you acces an array pos, it needs to get this array pos from ram. It is then stored in the cache, but right before it is used again, it is dumped. So the cache is essentially useless, leading to a much longer execution time.