AMD EPYC vs. Intel Xeon Cascadelake With Facebook's RocksDB Database
Written by Michael Larabel in Processors on 17 October 2019. Page 2 of 3. 9 Comments

Kicking things off with a random fill test, the previous-generation AMD EPYC 7601 2P performed similar to the dual Xeon Platinum 8280 server, so it's no surprise once moving to the new Rome processors their performance goes well beyond what is offered by Naples and Intel Cascade Lake CPUs tested. The Rome processors were delivering about 32% better performance.

When carrying out a random fill and forcing syncs to the Optane 900p disk in sync mode rather than async, the top-tier EPYC Rome configuration is 38% faster than the top-tier Intel Xeon Cascade Lake (non-AP) configuration.

When carrying out random reads to the RocksDB embedded database, the EPYC 7502 2P setup had a 8% advantage over the Cascade Lake CPU set tested while the EPYC 7642 CPUs led to a 48% win over Intel and the top-tier EPYC 7742 had a 66% advantage.

With sequential fills that do not scale out like other operations, the Rome CPUs still exhibited around 38% better performance than previous-generation Naples and Intel Cascade Lake.

Lastly was RocksDB with its read-while-writing test, which puts the EPYC 7502 2P at 19% faster than the Xeon Platinum 8280, the EPYC 7642 2P at 56% faster, and EPYC 7742 2P at 89% faster than the current Intel Xeon competition.

If looking at the geometric mean of all those different RocksDB operations, the tested EPYC Rome processors all delivered healthy advantages over the Xeon Platinum 8280 Cascade Lake dual socket server. The EPYC 7502 2P was 23% faster than the 8280 2P while having just 14% more threads (112 vs. 128 threads). AMD EPYC Rome is doing great with RocksDB and even compared to the previous-generation EPYC 7601 2P was a 30% uplift going to the EPYC 7502 2P or 1.6x the speed when going from top-tier Naples to top-tier Rome.


Related Articles
Trending Linux News