Windows vs. Linux Scaling Performance From 16 To 128 Threads With AMD Ryzen Threadripper 3990X

First up was looking at the NAMD performance. At 16 cores, the Windows performance came out slightly ahead of Linux with using the NAMD release builds on each operating system. At 32 cores the performance was similar between Windows and Linux, but at 48 cores and above is where Linux began offering much better performance than Windows 10 Enterprise and Windows 10 Pro. In the default Threadripper 3990X configuration of 64 cores + SMT (128 threads), the NAMD performance on Windows is much worse than 64 cores without SMT or even in the 48 core configuration. NAMD on Clear Linux continued scaling up through the 64 cores plus SMT.

Under the highly-threaded John The Ripper password cracker, the Blowfish performance scaled nicely on Linux when going from 16 cores/threads to 128 threads was a 4.4x speed-up under Clear Linux but just 2.5x on Windows 10 Pro/Enterprise. John The Ripper Blowfish was another case where the 64 cores plus SMT regressed the Windows 10 performance but Linux saw 1.29x the performance from enabling SMT. With this test, Windows 10 Enterprise was offering better performance over Windows 10 Pro.

The MD5 results out of John The Ripper were similar to the Blowfish algorithm with seeing great results out of Linux, Windows 10 dropping when hitting 64 cores + SMT, and Windows 10 Enterprise offering better performance each step of the way over Windows 10 Professional. At 16 cores, Windows 10 Enterprise had a slight lead over Clear Linux but at 64 cores, Linux was at 1.29x or 1.82x with the 128 thread setup.

GraphicsMagick has long utilized OpenMP threading for accelerating this image manipulation program. But on Windows the GraphicsMagick performance continues to scale terribly compared to the Linux performance. On Linux was at 2.8x from 16 to 128 threads.