NAS Parallel Benchmarks: EPYC 7601 vs. EPYC 7742 vs. Xeon Platinum 8280
Not included as part of our original EPYC 7742 / EPYC 7002 "Rome" Linux benchmarks was the NAS Parallel Benchmarks (NPB) developed by NASA. While an MPI testing favorite, there were build issues with the older version of NPB packaged by the Phoronix Test Suite. But with recently having updated that test profile against the latest NPB upstream, here are some results for the EPYC 7742 2P, EPYC 7601 2P, and dual Xeon Platinum 8280 benchmark results. Separately, there's also results now for NeatBench 5 with this video editing plug-in test case now part of the Phoronix Test Suite.
NPB 3.4 is now available via the Phoronix Test Suite / OpenBenchmarking.org as the current upstream release for these demanding benchmarks out of the NASA Advanced Supercomputing lab. Some of the larger test cases / data-set sizes are also enabled now to better deal with the increasing computing capacity from the likes of Cascadelake and Rome. All tests on Ubuntu 19.04 with Linux 5.3 Git kernel while running from Optane 900p NVMe SSD storage and each server having memory at its maximum number of memory channels and optimal frequencies.
With each EPYC 7742 processor offering 64 cores / 128 threads, the NPB results were quite dramatic compared to the previous-generation EPYC 7601 processors and Intel Xeon Platinum 8280 competition:
In some cases the EPYC 7601 2P does beat out the Xeon Platinum 8280 due to the greater number of cores/threads.
But in some cases where the previous-generation EPYC 7601 2P lost to the dual 8280s, the 7742 2P server now completely dominates over Cascadelake with these MPI benchmarks. In every NPB test run, the EPYC Rome server won compared to the Xeon Platinum performance.
If looking at the harmonic mean of these NPB results, the 7742 2P was 56% faster than the Xeon Scalable competition and 2.4x the speed of the previous-generation 7601 2P. (More data in this OpenBenchmarking.org result file.)
While unrelated to NPB, a new test profile recently added to the Phoronix Test Suite was NeatBench around the Neat Video video editing plug-in. So with these new/updated test profiles, I also happened to run NeatBench at the same time on these servers.
While the EPYC 7742 performance only came out slightly faster than the 8280 server, it was 3x the speed of the EPYC 7601 PowerEdge server. That's quite impressive for a generational difference. As far as why the EPYC 7742 doesn't have a larger lead over Intel, it appears NeatBench doesn't scale quite as well as other multi-threaded workloads and does have benefit on the second-generation Xeon Scalable due to the higher clock frequencies there. But with this being a new test profile, I am still evaluating this test more for its characteristics.
I'm still certainly very impressed by the performance potential of AMD EPYC Rome processors on Linux with the benchmarks going on now for nearly two months. Coming up next on the Rome benchmarking front is the long-awaited Linux distribution benchmark comparison.
Those wanting to see how their own system(s) compare to the results in this article can simply install the Phoronix Test Suite and run phoronix-test-suite benchmark 1909031-AS-1909030AS26.
NPB 3.4 is now available via the Phoronix Test Suite / OpenBenchmarking.org as the current upstream release for these demanding benchmarks out of the NASA Advanced Supercomputing lab. Some of the larger test cases / data-set sizes are also enabled now to better deal with the increasing computing capacity from the likes of Cascadelake and Rome. All tests on Ubuntu 19.04 with Linux 5.3 Git kernel while running from Optane 900p NVMe SSD storage and each server having memory at its maximum number of memory channels and optimal frequencies.
With each EPYC 7742 processor offering 64 cores / 128 threads, the NPB results were quite dramatic compared to the previous-generation EPYC 7601 processors and Intel Xeon Platinum 8280 competition:
In some cases the EPYC 7601 2P does beat out the Xeon Platinum 8280 due to the greater number of cores/threads.
But in some cases where the previous-generation EPYC 7601 2P lost to the dual 8280s, the 7742 2P server now completely dominates over Cascadelake with these MPI benchmarks. In every NPB test run, the EPYC Rome server won compared to the Xeon Platinum performance.
If looking at the harmonic mean of these NPB results, the 7742 2P was 56% faster than the Xeon Scalable competition and 2.4x the speed of the previous-generation 7601 2P. (More data in this OpenBenchmarking.org result file.)
While unrelated to NPB, a new test profile recently added to the Phoronix Test Suite was NeatBench around the Neat Video video editing plug-in. So with these new/updated test profiles, I also happened to run NeatBench at the same time on these servers.
While the EPYC 7742 performance only came out slightly faster than the 8280 server, it was 3x the speed of the EPYC 7601 PowerEdge server. That's quite impressive for a generational difference. As far as why the EPYC 7742 doesn't have a larger lead over Intel, it appears NeatBench doesn't scale quite as well as other multi-threaded workloads and does have benefit on the second-generation Xeon Scalable due to the higher clock frequencies there. But with this being a new test profile, I am still evaluating this test more for its characteristics.
I'm still certainly very impressed by the performance potential of AMD EPYC Rome processors on Linux with the benchmarks going on now for nearly two months. Coming up next on the Rome benchmarking front is the long-awaited Linux distribution benchmark comparison.
Those wanting to see how their own system(s) compare to the results in this article can simply install the Phoronix Test Suite and run phoronix-test-suite benchmark 1909031-AS-1909030AS26.
7 Comments