Intel Xeon Gold 5220R + Xeon Gold 6226R Linux Performance
At the end of February Intel launched the Xeon Scalable "Cascade Lake Refresh" processors with a number of more aggressively priced SKUs with different core counts and clock speeds compared to the original Cascade Lake CPUs launched last year. Intel recently sent over the Xeon Gold 5220R and Xeon Gold 6226R processors and we've begun our Linux benchmarks of them. In this article is our initial look at their performance using a near-final build of Ubuntu 20.04 and seeing how the performance stacks up in raw performance and performance-per-dollar against the AMD EPYC competition.
The Xeon Gold 5220R is a 24 core part plus Hyper Threading and features a 2.2GHz base clock with 4.0GHz turbo while having a 35.75MB cache. The Xeon Gold 5220R comes in at $1555 USD (though currently retailing for ~$1726 USD), which is priced much more competitively compared to previous Xeon CPUs.
The Xeon Gold 6226R meanwhile is a 16 core part plus hyper Threading with a 2.9GHz base clock and 3.9GHz turbo frequency while having a 22MB cache. The Xeon Gold 6226R comes priced at just $1300 USD (or about $1451 USD at current retail pricing), which is a big improvement in Xeon pricing.
Both of these Xeon Gold Cascade Lake Refresh CPUs have a 150 Watt TDP and can handle Intel Optane DC Persistent Memory support. The Cascade Lake Refresh processors offer up to 42% more performance per dollar and up to 36% more performance compared to 1st Gen Xeon Scalable.
Still being part of the Cascade Lake family, the Linux (and BSD) support isn't any issue at all. One of Intel's very strong selling points remain their superb open-source/Linux support at launch (and generally well ahead of launch). We are seeing AMD continuing to make significant strides in more punctual Linux support across their portfolio but there is still room for more improvement -- there still are even gaps around thermal reporting on Linux as well as the lack of CPU package power consumption reporting along with areas like AMD CPPC / CPUfreq enhancements and routinely the GCC / Clang compiler support for new functionality lagging. This timely Intel open-source support especially holds true on the BSD front with Intel CPUs and chipsets generally running fine across the various BSDs without any extra fuss.
This round of benchmarking was done with all CPUs freshly re-tested against Ubuntu 20.04 in its near final form with the Linux 5.4 kernel and GCC 9. All CPUs tested were running at their optimal memory frequency and maximum number of supported memory channels. Intel Optane 900p NVMe SSD storage was used throughout. The CPUs tested for this comparison included:
- Xeon Silver 4216
- 2 x Xeon Silver 4216
- Xeon Gold 5218
- Xeon Gold 5220R
- 2 x Xeon Gold 5220R
- Xeon Gold 6226R
- 2 x Xeon Gold 6226R
- Xeon Platinum 8280
- 2 x Xeon Platinum 8280
- EPYC 7302P
- EPYC 7502P
- 2 x EPYC 7252
- 2 x EPYC 7272
- 2 x EPYC 7302
- 2 x EPYC 7352
- 2 x EPYC 7532
A larger comparison will be out soon with still (re)testing the various CPUs using the new Ubuntu 20.04 base combined with the CPUs I have available. Via the Phoronix Test Suite we were running dozens of benchmarks across the tested CPUs. Continue on for the raw performance and performance-per-dollar metrics.