AMD Epyc Zen1 vs Zen2

Hello everyone,

I’m new to this forum. Perhaps this question has been answered before, but I could not find related topic for my specific needs.

I’m a student who completed his PhD in the field of numerical simulations.
As many of you know, using someone else’s code can be little (lot) frustrating. So I started dreaming of writing my own fluid simutation code.
The goal, for me, is not to write a professional, industrial code by myself, but it’ll help me on a lot of stages, from basic computer science to high level math and stuff.
I am interested in this as an enthusiast but I do not pretend being an expert (far from it).

To achieve my goald, I will use (I think) an MPI/OpenMP code and as you may know they require somewhat lots of cores.

I needed some advice, should I start with somethinh like a threadripper with 32 cores or go with an epyc processor with high bandwith?

If I’m missing some points, I’ll be happy to hear your advice.

Best wishes

What kind of load does

utilize, is it high in memory bandwidth or just raw compute?

Whats your budget?

I would start with looking at LLVM/Clang or GCC benchmarks.
Phoronix probably has some.

Other than that, I’m not sure how to recommend processor, if you’re not very sure what you will use it for…

Both are good processors. Depends on your budget.

EPYC and Threadripper are both very nice, but you haven’t even written your code yet.

If I were you I’d build a smaller system first with a Ryzen 3900X and then write some code. Once you find the bottlenecks using hardware profiling tools like Linux perf or whatever is available on Windows. AMD uProf? I don’t know if Intel VTune works on AMD. Anyway, once you discover if you need more cores or more RAM speed and how your code scales with both then you can decide.

Threadripper TR3/TRX40, is being slated for long term commitment. SP3 has a manaical chiplist, so you could get hands a lower core-count cpu, as baseline indicator of performance chutzpuh [if more becomes necessary], for the project.

Zen2 plays nicer, with the common I/0 dialogue position between all the chiplets, compared to the partial [2] chiplet bias of Zen1 [when concerning higher cored variants]

AMD_EPYC_die

Hi there,

I’m working as a post doc making large-scale vibroacoustic simulation and optimization codes. My code is mainly OpenMP based.

I would also agree you should maybe start out with a simple ryzen based computer. You can solve simple problems to start with, I would even start with a simple 2D implementation. Nevertheless, I personaly use an 18 core xeon and 256 GB ram.

Maybe out of curiosity, what do you plan to implement? A finite volume code usign k-omega or something more exotic?

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.