Intel's Linux Software Optimizations Still Pay Off For Xeon 6700E "Sierra Forest" E-Core CPUs
When testing Intel's aggressive software Linux optimizations shipped by way of their in-house Clear Linux distribution, I am most often testing it on their high core count Xeon processors with AVX-512... Over the years in dozens of Phoronix articles there have been countless metrics showing off the out-of-the-box performance benefits from leveraging software built for higher x86_64 micro-architecture feature levels, employing compiler-based function multi-versioning, and the other extensive performance tuning carried out by Intel software engineers. But now with the Intel Xeon 6700E "Sierra Forest" series now being available for these all-E-core server processors, I was curious about quantifying the Clear Linux benefits over the likes of Ubuntu Linux. Here are those benchmarks for those curious about the difference.
Intel's Clear Linux platform is extensively tuned for AVX-512 and other modern CPU ISA features, but thanks to the likes of GCC FMV and x86_64 micro-architecture feature level handling by Glibc, AVX-512 isn't mandated. In fact, Clear Linux continues to advertise support for Intel Core 2nd Gen and newer, Intel Atom CPUs, and Intel Xeon E3 / E5 / E7 processor support. SSE 4.2, SSSE3, and PCLMUL are among their baseline requirements. But still I was wondering about the performance benefits to the E-core-based Intel Xeon 6700E Sierra Forest platforms with Clear Linux relative to the more common Ubuntu Linux. Plus I also tossed in Arch Linux as another popular rolling-released based Linux operating system.
On the Intel Xeon 6700E reference platform with two Xeon 6780E processors providing a combined 288 cores and 512GB of RAM and a Samsung MZWLJ1T9HBJR-00007 NVMe SSD, I ran the following Linux OS tests:
- Clear Linux 41900 as the latest of Intel's in-house Linux distribution as of testing. Clear Linux 41900 employs the Linux 6.9 kernel, GCC 14.1.1 compiler, EXT4 by default, and other up-to-date packages.
- Arch Linux for that rolling-release that as of testing was using Linux 6.9, GCC 14.1.1, and other up-to-date software versions largely similar to Clear Linux. For some sense of defaults, Archinstall was used for setting up Arch Linux on the Sierra Forest server with its defaults.
- Ubuntu 24.04 LTS out-of-the-box with the Linux 6.8 kernel, GCC 13.2 compiler, and other defaults.
- The Ubuntu 24.04 LTS benchmarks were also repeated when switching over to the P-State "performance" governor. Among many other changes made to Clear Linux in the name of greater performance, it uses the "performance" governor by default rather than P-State's default "powersave". The performance governor is typically used within my server CPU reviews and the like and other server benchmarks and frankly wish all the server Linux distributions would default to the performance governor... Anyhow, for ruling in/out the workloads where the CPU frequency scaling governor makes a significant difference, this secondary "Perf Gov" data-set is the same Ubuntu 24.04 but switching over to the P-State driver's performance governor.
From there many different workloads were carried out while also monitoring the combined CPU power consumption of these Xeon 6780E Sierra Forest processors during testing. Let's see how Intel's Linux software optimizations are still very beneficial even for E-core-only Sierra Forest.