CPU related question: More cores or higher clock speed?

[HansWellens]

I know Zephyr is Cuda accelerated whenever possible. So the number of Cuda cores (and cards) is probably the most important factor there.
But what about the processor? Is Zephyr's processing speed determined by the number or cores (or threads), or is (single) core speed more important?
Are there architectural differences between Intel or AMD in the way they are more or less suited for processing acceleration?

[Andrea Alessi]

hi Hans,

AMD and Intel shares the same x86_64 instruction set so there are no architectural differences in that regard. We do use SSE instruciton optimizations and both AMD and Intel processors support them - so there is no preference in either one.

Zephyr also uses as many threads as available, however certain specific parts of the algorithm are not - by nature - parallel, so in some specific parts you may see a lot of your processor not being used. What is best? It's really hard to tell, it depends on the type of datasets and settings. To be honest however, if you have at least an NVIDIA GPU, you offload most of the workload to the video card so a safe ballpark when evaluating the best cpu is simply looking at its clock / cache / cost.

[cam3d]

Good to know Andrea! - I was under the impression that there was an Intel leaning preference. I'm looking at new chips right now... 7820x is calling! Mighty upgrade from my 2600k