Benchmark on AMD EPYC 7T83 and INTEL XEON 8375C
 

Recently I ran some benchmarks on the 2*7T83 and 2*8375C platforms. The benchmark case was a transient CFX analysis with 21.33M Elements and 4.05M Nodes.

Besides the CPU and MB, other specs are identical in both platforms (1T NVME SSD, 512GB DDR 3200 ECC RAM, etc...)


Results can be found in the attached figure. The 2*7T83 platform is 50% faster than its 2*8375C counterpart due to doubled cores. But the price of 2*7T83 (platform price around 12K USD) almost also doubled.

I'm wondering if a cluster with 2 nodes of 2*8375C (connect with 100GB IB) would catch up with a single node with 2*7T83:D
https://www.cfd-online.com/Forums/me...layout-2-1.jpg

https://www.cfd-online.com/Forums/me...6-layout-3.jpg

In case that's an actual question: yes, very likely. For only 2 nodes, you probably don't even need fancy Infiniband to get decent scaling. Ethernet might be enough.

Thanks! I'm still swinging between 2 nodes of 8375C and 1 node of 7T83(OEM version of 7663). I'll try to carry some tests