Benchmarks

It has always been a question of mine: What is the effect, if any, of  turning Simultaneous Multi-Threading (SMT) on/off in AMD processors for the performance of the Self-Initializing Quadratic Sieve (SIQS)?

I tested on a 2 Socket System with 2x AMD EPYC 7642 on the same motherboard. The results is the image on the left. Notice, it is a log2 scale on the y-axis. To really see the difference. See the next image.

It is hard to see how the OFF/ON runtime compare on a log2 scale. Here's a clearer picture of they compare. Before about 310-bits or so, SMT turned ON is about the same or better, but after 310-bits or so SMT turned off starts to do significantly better. 

By the time we get to 360-bits it does 3X better, that is to say, SMT OFF runs in 1/3 the time as does YAFU SIQS with SMT turned on. It seems this is the peak, It starts to level off after that.