POWER9 & ARM Performance Against Intel Xeon Cascadelake + AMD EPYC Rome
This round of testing is a more limited sub-test than our other EPYC vs. Xeon benchmarks due to limiting the tests that have decent exposure/potential on POWER9 and ARM. With Rodinia as an example, the POWER9 performance is competitive to the x86_64 CPUs while the ARMv8 performance is a great deal behind.
The Java/OpenJDK benchmarks obviously work well cross-platform like DaCapo Bench. Though in the case of this Jython sub-test it's largely single-threaded performance and thus the Ampere eMAG with its 3.3GHz clock speed does much better than Cavium's cores at 2.0GHz. The POWER9 Talos II was the most competitive in this instance in not being too far behind the x86_64 competition.
Though in other Java conditions, the Ampere and Cavium servers did have a better showing than POWER9 but were no match to the AMD Rome or Intel Cascadelake performance.
With x264 as a video encoding test, the POWER9 and Arm performance was a great deal behind the x86_64 CPUs. We're seeing more video encoders begin to optimize for ARM and POWER9, but with so much of the performance sensitive code-paths coming down to hand-tuned Assembly, it's to little surprise seeing EPYC and Xeon performing much faster in x264 and others.
When running the 7-Zip compression benchmark, the POWER9 Talos II with its 176 threads pulled in between the previous-generation EPYC 7601 2P and dual Xeon Platinum 8280.