Intel launched its Xeon Sapphire Rapids Scalable processors the other day after what can only be described as a generational delay. The core counts have been scaled up to 60 courtesy of a modular design, and pricing is comparable to rival Epyc offerings, but there aren’t many scenarios that can be counted as wins. A set of benchmarks from ServeTheHome show that Intel’s 4th Gen Xeon chips barely manage to compete with AMD’s last-gen Epyc Milan offerings:
The above Linux kernel compilation test shows the fastest Xeon Sapphire Rapids-SP CPU (60 cores) levels with the Epyc 7773X, a 64-core Milan part released a year ago. The Zen 4-based Genoa offerings are in a league of their own. The dual Epyc 9554 (64 cores per socket) is 25-30% faster than the top Intel combo, while the Dual 9654 (96 cores per socket) is over 70% faster than the Sapphire Rapids-SP flagship.
Kernel-based Virtual Machine (KVM) allows administrators to create multiple virtual and discrete environments (sessions/logins/portals/shared server, etc.) using a single machine. It’s part of Linux 2.6 and higher and is a key data center workload.
The Dual Epyc 9654 setup with 1,536GB DDR5 memory is around 70% faster than the Xeon Platinum 8490H combo. It’s a clash of two 96-core against two 60-core SKUs with different memory configurations. However, the fact that the 64-core Epyc 9554 pair is 25% faster than the latter makes it hard to defend the Intel offering. The Xeon Platinum 8490H mostly competes with the last-gen Milan and Milan-X flagships.
Similar results can be seen in MariaDB and Nginx. The Genoa parts are faster, with a lead of 10% to 40%, while the 64-core Milan-X deals with Sapphire Rapids. If you look back at the original launch schedule of the 4th Gen Xeon family, these results aren’t surprising. Originally, Sapphire Rapids was built to compete with Milan and Milan-X, which it does quite well. However, due to the repeated delays, it has to put up with the more advanced Genoa lineup.
Thanks to STH for the thorough testing. Cheers!