Iman Hosseini

I do not have an application in mind, just wanted to help with the issue. The numbers are total numbers for 10 runs (which is what the test prints out...

Using horner, now it's better then method2:  And concentrating more on low N regime for fixed M:  And for scaled M (M = 2(N1xN2xN3)):  An interesting view...

I guess that A. the crossover is gpu-dependent. and B. also dependent on N. Here is what I have from 40x40 data points for various N, M/8N: (on 3060 Ti)...