Processor multi core speed question - user benchmark
 

Hello,

User benchmark is a site that collects data from thousands of users online and let's you compare different processors according to many different parameters.
https://cpu.userbenchmark.com/

Some of those parameters of interest include the 1 core-speed,2-core speed, 4-core speed, 8-core speed and the 64-core speed.

According to user benchmark n-core speed "... measures the ability of a processor to perform n integer and floating point operations at the same time."

I'm doing 2D simulations, but my fx 6300 is considerably too slow for what I feel comfortable with, given the type of 2D simulations that I am doing. I want to invest in a pc and better processor. For example Ryzen 9 3950x.

Here is the comparison between the two.
https://cpu.userbenchmark.com/Compar...300/4057vs1555

If the 64-core speed was one that mattered then I would achieve probably about 7-8 times speed up.

My guess is that a processor doing a CFD simulation would want to do as many floating point operations at the same time as possible, hence 64-core speed would be most representative metric of processor speed for CFD. But then, my understanding of the issue is rather shallow.

That's why I would like to ask, if the 64-core speed the metric I should use for comparing processors working on a CFD simulation or perhaps lower core speed would be more representative?

I'm solving for mass, momentum and energy in each cell in URANS type simulation.

Disclaimer : Not an expert

Not a solution to your question, but something to be aware of.
Make sure your code scales almost linearly with the increase in processor count.
I have seen codes that don't scale linearly with increase in processor count, because they were not written with scalability in mind.
In such cases, buying a high cost CPU didn't lead to expected improvements in performance.
This generally happens in research codes. i.e you're writing your own code, or you got it from your professor.
If you're using a commercial code, you're most probably okay, and don't have to worry about scalability.

Most of the popular benchmarks you find online are not a good estimate for CFD performance. Userbenchmark in particular has some other problems, but let's not go down that rabbit hole.
The main issue is: they focus on aggregate floating point or integer performance, which is only one part of the picture for CFD - and many other "real-world" applications for that matter.

shameless plug of my latest work:
https://www.cfd-online.com/Forums/ha...dware-wip.html