Intel Xeon Platinum 8380 Ice Lake Linux Performance vs. AMD EPYC Milan, Cascade Lake

Out of the 110 tests carried out in total across these server processors with a very diverse range of workloads, the Xeon Platinum 8380 came out to be 31% faster than its predecessor, the Xeon Platinum 8280 Cascade Lake. While hefty generational lift, overall this puts the 8380 2P in line with the dual EPYC 7F53 performance. Overall, the EPYC 7763 came out to be about 10% faster than the dual 8380 configuration across all of the benchmarks.

When it came to the code compilation performance (measured across timed Linux kernel, ImageMagick, LLVM, FFmpeg, Build2, Godot, and WASMer) builds, the Ice Lake server was performing between the EPYC 7713 and 7763 processors. It ultimately comes down to what code-base(s) you would be building a lot, their build system and the number of source files, and other factors for how competitive Ice Lake is against the EPYC 7003 series and whether the extra cores from AMD will be beneficial.

For HPC benchmarks (counted as NPB, Rodinia, HPCG, Quantum Espresso, FFTE, MT-DGEMM, AMD, NAMD, GROMACS, NWChem, LAMMPS, Lulesh, OpenFOAM, Caffe, Numpy, Numenta NAB, TensorFlow Lite, oneDNN, OpenVINO, and Lczero), the Xeon Platinum 8380 dual socket server led with about 4% better performance than the EPYC 7763.

The Xeon Platinum 8380 2P performed particularly well in the machine learning workloads thanks to many of them being optimized for Intel / AVX-512.

The other area where the 3rd Gen Xeon Scalable processors were especially strong are with Intel's own oneAPI software packages, which to little surprise are well optimized for Intel's latest wares. OneDNN, Open Image Denoise, OSPray, OpenVKL, and OpenVINO are among the real-world open-source Intel software packages we routinely run among our benchmarks.

If looking just at the molecular dynamics workloads, the EPYC 7763 came out as the stronger processor.

Or if looking at all of the tests making use of OpenMPI, the Xeon Platinum 8380 comes out slightly ahead of the EPYC 7713.

AMD EPYC 7003 series leads with ease when it comes to CPU-based rendering performance with the likes of Blender, OSPray, and LuxCoreRender where they scale well with increasing core counts. But at least from Cascade Lake to Ice Lake is a big improvement.

AMD also led with the video encoding workloads where the greater core counts found with AMD help out especially for 4K content.

If looking at the single threaded performance of the Xeon Platinum 8380 in workloads including Botan cryptography, Python, PHP, and HINT, the Xeon Platinum 8380 came out ahead of the EPYC 7713/7763 processors but behind the frequency-optimized EPYC 75F3.

Generationally there is significant uplift going from the Xeon Platinum 8280 to Xeon Platinum 8380, but how competitive that puts 3rd Gen Xeon Scalable against the AMD EPYC 7003 series competition largely comes down to the particular workload(s) of interest/relevance to you as shown by this wide variety of testing. Those wanting to go through all individual benchmarks along with per-test power efficiency data in full can view them on OpenBenchmarking.org.

Additional benchmarks looking at the Intel Xeon Platinum 8380 performance across other Linux distributions, the BSD/FreeBSD/DragonFlyBSD support and performance, GCC vs. LLVM Clang compiler impact, and other interesting areas of performance for Ice Lake are among the upcoming articles as we continue exploring 3rd Gen Xeon Scalable performance. As always, if there are any workloads not covered by our Phoronix Test Suite / OpenBenchmarking.org testing that can automate well and meet my other test requirements, I am always interested in adding those tests - let me know via the forums or Twitter, to complement the 600+ benchmarks in total available via our open-source benchmarking software. Thanks to Intel for providing the Ice Lake reference server being used for this benchmarking.