Intel’s EPYC Benchmarks Shine A Light On Intel’s Strengths Albeit With Potential Caveats
Intel recently released a slide deck which showcases performance metrics tested deep in Intel Labs. While we would always advise you to take any first party metric with a pinch of salt, Intel does appear to be reasonably reserved in taking unnecessary pot shots at AMD like they have done in the past, and sticking to the facts.
Intel’s EPYC benchmarks showcase impressive performance metrics, but at a premium
The slide deck compares a fully tricked out a 2S EPYC 7601 server with a 2S Intel Xeon 8160. Both systems cost roughly the same ($16128 for the Intel one vs $16280 for the AMD one). While this might seem counter intuitive, considering Intel parts are priced higher, the difference can be explained with the differences in board pricing as well as lower sticks of RAM (the AMD variant uses 16 sticks of DDR4 while as the Intel variant uses 12). Both server racks are from Supermicro.
Here’s the catch though, this particular setup uses an Intel 8160 part in 2S mode, whiles the benchmarks also showcase the Intel 8180 in all its glory, which unlike the former is $10,000 part. This means shifting from the 8160 to the 8180 you are looking at a price difference of roundly $12000 for a 2S configuration. It will do well to keep that in mind while absorbing the following benchmarks.
Here we see that the Intel 8160 just barely outperforms the competition and is more or less neck to neck with the EPYC offering. Since they are both in the same price range, this is the more relevant comparison over here. The 8180 is much further ahead then the EPYC 7601 rack and is clearly in a league of its own.
Intel hasn’t missed its chance to point out the bandwidth issues that are present in die on the EPYC processors. Because of the fact that it uses an MCM approach to building a processor, there are inherent latency issues. In conditions likes these, Intel’s monolithic approach is superior. But the question is, does this actually translates to a real life difference in performance? This is a question that has yet to be answered since we don’t have a reference AMD monolithic design of these proportions to compare against and it is fairly competitive against Intel offerings.
Intel has also summarized using the SPECint metric and given a value of the performance per core that you are getting from all three server racks. As expected the Intel offerings do offer superior performance to AMD counterparts. There are several caveats associated with this slide deck though. While I am fairly certain the benchmarks themselves can be relied on, keep in mind that Intel did not use the GCC compiler (and really there is no logical reason for them to do so). This is perfectly fine considering customers wouldn’t really use a GCC compiler instead of Intel’s own.
Secondly, Intel did not include power efficiency metrics which are very important in themselves and can dictate server decisions just as much as as cost. And finally of course, the price tag. Intel’s performance with the 8160 is slightly better than the EPYC 7601 and comes at a competitive pricing, the 8180 however, is priced at a much higher premium – but does deliver the absolute best in terms of performance as well, to be fair.