Intel Xeon P-State vs. CPUFreq Scaling Tests Show Some Odd Numbers
As we've seen a lot of variation in results with different Intel processors when switching between the Intel P-State and CPUFreq scaling drivers and the different governors, here's some tests when using a 16 thread (eight core + HT) Haswell-EP Xeon processor and testing the different CPU frequency scaling settings in Fedora 21.
If you haven't seen our earlier CPUFreq vs. P-State articles of recent time, first read Intel P-State vs. CPUFreq Benchmarks On The i7-5960X, Ondemand vs. Performance CPU Governing For AMD FX CPUs, and ACPI CPUfreq vs. Intel P-State Scaling With Linux 3.15 (there's also numerous other articles on the topic at Phoronix you can search for if wanting to learn more). The purpose of today's article is to pass along some new numbers, with the primary focus of testing a Xeon server/workstation CPU rather than a desktop CPU as used in past tests.
The Intel Xeon E5-1680 v3 Haswell-EP processor has a stock frequency of 3.2GHz with a Turbo Boost of 3.8GHz. During testing the CPU settings were left at their defaults. The motherboard used was the MSI X99S SLI PLUS and the Xeon CPU sample was also kindly provided by MSI. A development snapshot of Fedora 21 x86_64 was running on the system with the Linux 3.18-rc3 Git kernel. The CPU frequency scaling conditions tested were P-State Powersave (the system default), P-State Performance, CPUFreq Ondemand, CPUFreq Performance, and CPUFreq Conservative.
In the OpenGL tests ran with a GeForce GT 740 on Nouveau, there wasn't too much of a change in performance except at extreme settings.
For compiler workloads like compiling the Linux kernel, the results were particularly odd for this CPU.... CPUFreq Ondemand and CPUFreq Conservative ended up being significantly faster than other conditions -- including the performance governor with both P-State and CPUFreq. This result is particularly odd.
The FFmpeg result looks like how it should be roughly with regard to the governor performance differences.
Find more of these CPU scaling benchmarks for the Intel Haswell-EP Xeon at OpenBenchmarking.org and stay tuned for more Intel Xeon Linux benchmarks shortly on Phoronix.
If you haven't seen our earlier CPUFreq vs. P-State articles of recent time, first read Intel P-State vs. CPUFreq Benchmarks On The i7-5960X, Ondemand vs. Performance CPU Governing For AMD FX CPUs, and ACPI CPUfreq vs. Intel P-State Scaling With Linux 3.15 (there's also numerous other articles on the topic at Phoronix you can search for if wanting to learn more). The purpose of today's article is to pass along some new numbers, with the primary focus of testing a Xeon server/workstation CPU rather than a desktop CPU as used in past tests.
The Intel Xeon E5-1680 v3 Haswell-EP processor has a stock frequency of 3.2GHz with a Turbo Boost of 3.8GHz. During testing the CPU settings were left at their defaults. The motherboard used was the MSI X99S SLI PLUS and the Xeon CPU sample was also kindly provided by MSI. A development snapshot of Fedora 21 x86_64 was running on the system with the Linux 3.18-rc3 Git kernel. The CPU frequency scaling conditions tested were P-State Powersave (the system default), P-State Performance, CPUFreq Ondemand, CPUFreq Performance, and CPUFreq Conservative.
In the OpenGL tests ran with a GeForce GT 740 on Nouveau, there wasn't too much of a change in performance except at extreme settings.
For compiler workloads like compiling the Linux kernel, the results were particularly odd for this CPU.... CPUFreq Ondemand and CPUFreq Conservative ended up being significantly faster than other conditions -- including the performance governor with both P-State and CPUFreq. This result is particularly odd.
The FFmpeg result looks like how it should be roughly with regard to the governor performance differences.
Find more of these CPU scaling benchmarks for the Intel Haswell-EP Xeon at OpenBenchmarking.org and stay tuned for more Intel Xeon Linux benchmarks shortly on Phoronix.
8 Comments