Over the last few months we have built two new workstations, one with a Intel 9900K, and the other with a AMD 3900x. We have made a report (see attached) that details our findings when solving Flow simulation files on these systems. Puget Systems also did testing with these two processors as well https://www.pugetsystems.com/labs/articles/SOLIDWORKS-2019-SP3-AMD-Ryzen-3-vs-Intel-9th-Gen-Core-1555/
In Puget's testing they found that the Intel 9900K was either equal or slower than the AMD. We only found the opposite to be true, that the Intel was always faster. Also we did further testing than Puget, by testing at a per core basis which showed that there was only a tiny bit to be gained by utilizing hyper threading and going over 8 (of 16) logical processors on the Intel, and nothing to be gained by going over 10 (of 24) logical processors on the AMD. However this did not hold true if you were doing multiple parallel simulations at once, which did benefit more with hyper threading. With only one simulation running, one core of the Intel reached 1x performance. With all the cores being utilized the Intel got 4.5x this performance, but the AMD only got a 2.5x performance. This was further increased to 7.5x (Intel) performance and 7x (AMD) performance when 8 simulations were run at the same time in parallel. Considering that the Intel is an 8 core, the theoretical performance for these parallel simulations is 8x, and the AMD should 12x, but this was not what was found.
After showing this to our VAR which contacted Solidworks directly they suggested to disable hyper threading all together on the AMD, which we have yet to test.
Given that our VAR only suggests more cores for a computer build, and that the AMD has better IPC and higher core count than the Intel, why is Flow simulation still so much faster on the Intel?