Is there a speed improvement opportunity in threading for the CPU only model?

[RndyP]

I've been using Whisper for a couple years now with my 4 core Intel box. 4-6 is the "sweet spot" for number of threads. I just got a new box with 20 cores. Since we get diminishing returns, I do not see that much of an improvement. Is there something that could be improved with the threading architecture? The builds that use GPUs are much faster. Am I to assume that the threading model is different due to the fact GPUs have many cores? I've looked at the ggml code a bit, but it's way over my head, so someone with much greater knowledge than me would have to take a look at it. When I use Resource Monitor in Windows, my 20 cores are hardly being used.

I might add that I use the Parallel Patterns Library (PPL) in other projects, it makes things really easy....