The Performance Impact Of Turbo Boost On Linux
After the tests this weekend of Intel Hyper Threading and core scaling with the new Intel Xeon E3 "Skylake" processor, I proceeded to do some Turbo Boost benchmarking.
Namely, I ran a set of open-source Linux benchmarks when the Xeon E3 1280 v5 was running in its default mode with four cores plus Hyper Threading and Intel EIST+TB enabled and then again when disabling Turbo Boost support from the BIOS. The Xeon E3 1280 v5 has a 3.7GHz base frequency and 4.0GHz boost frequency. The 80 Watt TDP processor was cooled by a Arctic Freezer i11.
The default P-State CPU scaling driver was in use for the E3-1280v5 and the performance governor was enabled.
While the CPU should have been in its turbo state for most of the benchmarking process as the cooling was more than adequate and the CPU core temperature was monitored and very low, there wasn't any difference when disabling Turbo Boost:
The results barely budged when disabling the Turbo Boost from the BIOS. You can see all of the data via this OpenBenchmarking.org result file. Have you encountered any similar issue under Linux? Is Turbo Boost somehow still getting enabled even if disabled from the BIOS or is TB actually so rarely being used with these higher-frequency processors? I will run the tests again with MONITOR=cpu.freq set so the Phoronix Test Suite will capture what frequencies are actually being used during the benchmarking process. I may also hook up the WattsUp Pro to it as well, plus maybe trying one one of the other 9 Intel Skylake E3 CPUs being benchmarked, but just figured I'd share this as a quick Sunday post to see if anyone had any immediate insight or similar encounter.
Namely, I ran a set of open-source Linux benchmarks when the Xeon E3 1280 v5 was running in its default mode with four cores plus Hyper Threading and Intel EIST+TB enabled and then again when disabling Turbo Boost support from the BIOS. The Xeon E3 1280 v5 has a 3.7GHz base frequency and 4.0GHz boost frequency. The 80 Watt TDP processor was cooled by a Arctic Freezer i11.
The default P-State CPU scaling driver was in use for the E3-1280v5 and the performance governor was enabled.
While the CPU should have been in its turbo state for most of the benchmarking process as the cooling was more than adequate and the CPU core temperature was monitored and very low, there wasn't any difference when disabling Turbo Boost:
The results barely budged when disabling the Turbo Boost from the BIOS. You can see all of the data via this OpenBenchmarking.org result file. Have you encountered any similar issue under Linux? Is Turbo Boost somehow still getting enabled even if disabled from the BIOS or is TB actually so rarely being used with these higher-frequency processors? I will run the tests again with MONITOR=cpu.freq set so the Phoronix Test Suite will capture what frequencies are actually being used during the benchmarking process. I may also hook up the WattsUp Pro to it as well, plus maybe trying one one of the other 9 Intel Skylake E3 CPUs being benchmarked, but just figured I'd share this as a quick Sunday post to see if anyone had any immediate insight or similar encounter.
21 Comments