A Curious Look At Eight Core Server CPU Performance From Intel Xeon Haswell To AMD EPYC Rome

When it comes to the AMD EPYC 7002 "Rome" processors we have looked at the various higher-end SKUs since their launch last August up to and including the EPYC 7742 with its 64 cores / 128 threads per socket. But for those wondering about the EPYC 7002 series performance at the bottom end of the spectrum, here are some fun benchmarks of the EPYC 7232P and EPYC 7262 on the near-final Ubuntu 20.04 LTS state compared to various vintages of Intel Xeon CPUs -- most notably, a curiosity driven look at the 8 core / 16 thread Intel Haswell Xeon performance.

The EPYC 7232P comes in at the bottom end of the EPYC spectrum at around $500~509 USD for the retail price as an 8 core / 16 thread Zen 2 server processor. The EPYC 7232P has a 3.1GHz base clock and 3.2GHz boost clock while having a 32MB L3 cache and a 120 Watt TDP. Common EPYC Rome features like the eight channels of DDR4-3200 remain supported with this low-priced EPYC processor.

The EPYC 7262 meanwhile at about $650 USD is still 8 cores / 16 threads but supports running in a 2P configuration should you want to do so now or in the future. The EPYC 7262 has a 3.2GHz base clock and 3.4GHz boost clock while having 128MB of L3 cache and a 155 Watt TDP.

For going with these lower-cost EPYC SKUs, the tests were done using the ASRock Rack EPYCD8 as one of the most budget friendly EPYC Rome 1P motherboards currently in the market.

Both of these processors were tested in a 1P configuration against various other CPUs I had available. This article is mainly intended for providing some reference benchmark figures off a near-final snapshot of Ubuntu 20.04 with that Ubuntu LTS release coming out in April. This isn't a complete review with not having access to any similar Xeon 8c/16t Cascade Lake processor but for those that may be running older 4c/8t or 8c/16t servers, hopefully you find these metrics interesting with the assortment of hardware I did test off Ubuntu 20.04.

EPYC 7251 - The former Naples 8c/16t EPYC CPU that has a 2.1GHz base clock and 2.9GHz boost clock with 32MB L3 cache and 120 Watt TDP.

Xeon E5-1680 v3 - For those that may still be running once powerful Haswell servers, the Xeon E5-1680 v3 is an 8 core / 16 thread part with 3.2GHz base frequency and 3.8GHz turbo while having a 20MB cache and a 140 Watt TDP. Back in the day this CPU retailed for $1700+ while now AMD's latest CPUs like the EPYC 7232P are retailing for just over $500. Worth noting is that the 3.2GHz base clock matches that of the EPYC 7262. This E5-1680 v3 is the primary point of interest for today's comparison in seeing how 8c/16t CPU performance compares for Haswell against Rome.

And then some other Intel Xeon 4c~8c parts tossed in for added reference:

Xeon E5-2687W v3 - While not an eight core part, this Haswell server CPU has 10 cores / 20 threads with a 3.1GHz base frequency and 3.5GHz turbo with a 25MB cache and 160 Watt TDP. This CPU launched in 2014 at above $2100 USD.

Xeon E5-2609 v4 - A Broadwell 8-core CPU without Hyper Threading and a 1.7GHz base frequency without any turbo. This low-end Broadwell server CPU has a 85 Watt TDP and 20MB cache.

Xeon E3-1275 v6 - As slightly newer, the Xeon E3-1275 v6 Kabylake processor with 4c/8t and 3.8GHz base frequency with 4.2GHz turbo and a 75 Watt TDP.

Each of the systems had memory at the maximum rated frequency supported and the optimal number of memory channels. The same SATA 3.0 Crucial SSD storage was used on all systems tested. It's unfortunate no Cascade Lake 8c/16t part for comparison in this article but at least interesting for seeing how the 8c/16t Haswell performance compares to 8c/16t of EPYC today while all of this testing was done in the past few days off an Ubuntu 20.04 snapshot with the Linux 5.4 kernel. In particular these results will hopefully help out if you wait 5+ years between any low-end/SOHO server upgrades. Power consumption numbers weren't used due to the varying motherboards and other components at play due to the span of hardware platforms tested.

These benchmarks were conducted via the Phoronix Test Suite. If you would like to see how your own Linux system(s) compare to the benchmarks about to be presented, you can conduct your own side-by-side, fully-automated benchmark comparison by running phoronix-test-suite benchmark 2003275-NI-EPYC7262L37.