Benchmarking The Intel P-State, CPUfreq Changes

On Friday there was the controversial news about the Linux "ondemand" cpufreq governor no longer being fit for best performance and power-savings on modern processors. Fortunately, for better handling the CPU frequency stage changes on modern Intel CPUs, Intel recently introduced the new P-State kernel driver.

With this news, plus word that changing the cpufreq governor can really boost the Mesa performance, many Phoronix readers were excited with 3+ pages of comments.

Before getting too excited, however, this doesn't mean immediate gains for all systems. The first system I ran some tests on, in fact, yielded no real performance changes for Sandy Bridge. This system was an Apple Mac Mini sporting an Intel Core i5 2415M "Sandy Bridge" processor and was running Ubuntu 13.04.

Tested configurations were using the Linux 3.8 kernel as found in Ubuntu 13.04 when changing out the CPU frequency governor from the ondemand default to performance, conservative, and powersave modes. Additionally, a Linux 3.10 Git kernel was tested with its frequency governor modes of powersave and performance. This 3.10 testing was from the new Intel P-State driver, which requires the kernel be built and can be checked if it's used at run-time through the scaling_driver sysfs attribute to see it's the P-State driver and not ACPI cpufreq.

For those unfamiliar with changing the CPU frequency governor, some documentation to look at includes the Fedora CPUfreq setup and Debian.org Wiki although the steps are rather distribution agnostic.

This initial testing from the Intel Sandy Bridge system can be found on OpenBenchmarking.org inside 1305183-UT-INTELLINU29.
Overall, these Intel Linux OpenGL results aren't too exciting when comparing the different governors and kernels. However, further testing is currently being done from Sandy Bridge and Ivy Bridge hardware -- including from mobile systems to measure the power consumption via the Phoronix Test Suite -- and from a more diverse variety of open-source Linux benchmarks. Stay tuned for more in the coming week and any questions/suggestions can be directed to Twitter.