NACK/Cmt: [SRU][N][PATCH 0/5 v2] Update default Intel EPP Values

Thibault Ferrante thibault.ferrante at canonical.com
Mon Sep 9 12:14:08 UTC 2024


Non-existing upstream commits listed in each commit sections.
Discussed via back channel.

On 04-09-2024 16:07, Pedro Henrique Kopper wrote:
> BugLink:  https://bugs.launchpad.net/bugs/2077470
> 
> SRU Justification:
> 
> [ Impact ]
> 
>   * Intel currently provides an Energy Performance Preference (EPP) tunable via its intel_pstate
>     to control power management. It ranges from 0 (performance) to 255 (powersaving).
>   * When the governor is set to powersave, which is the default, the EPP value of balance_performance
>     controls how the CPU frequency scaling occurs. However, on some platforms (Emerald Rapids, Meteor
>     Lake, Arrow Lake) it is set to a very conservative value out of the box, resulting in degraded
>     performance by default.
>   * These EPP tuning values work as a black-box and vary per-generation, so having the latest values
>     available is important to keep systems running at maximum achievable performance. Both Intel and
>     Canonical perform this kinds of tests and supply values.
> 
> [ Test Plan ]
> 
>   * Perform a benchmark on a stock Ubuntu 24.04 image
> 
>   * On Emerald Rapids:
>   * Change the EPP to 32 by executing
>     echo 32 | sudo tee /sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference
> 
>   * Observe the performance difference. In our internal testing, we found the following results
>     when setting EPP to 32:
>     POV-Ray: 32% faster | 12% less energy
>     OpenSSL: 12% faster | energy within 1%
>     Build Linux Kernel: 29% faster | 18% less energy
> 
>   * On Meteor Lake and Arrow Lake:
>   * Change the EPP to 64 by executing
>     echo 64 | sudo tee /sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference
> 
>   * Observe the performance difference. Intel found out the following for Meteor Lake:
>   > Speedometer 2.1
>   > score: +19%
>   > Perf/watt: +5.25%
>   >
>   > Webxprt 4 score
>   > score: +12%
>   > Perf/watt: +6.12%
>   >
>   > 3DMark Wildlife extreme unlimited score
>   > score: +3.2%
>   > Perf/watt: +11.5%
>   >
>   > Geekbench6 MT
>   > score: +2.14%
>   > Perf/watt: +0.32%
>   >
>   > Also update balance_power EPP default to 179. With this change:
>   > Video Playback power is reduced by 52%
>   > Team video conference power is reduced by 35%
>   * The same outcome should be obtained from Arrow Lake as per Intel.
> 
> [ Where problems could occur ]
> 
>   * The regression risks are very low. The CPU should perform in a more efficient zone, resulting in
>     overall power savings, despite the more aggressive scaling. This was verified using a Yokogawa WT310
>     power meter and an Emerald Rapids reference platform.
>   * Tests on Meteor Lake and Arrow Lake were performed by Intel. They were not independently verified by
>     Canonical.
> 
> [ Other Info ]
> 
>   * Patch for Emerald Rapids: https://lore.kernel.org/lkml/Zqu6zjVMoiXwROBI@capivara/ (merged in 6.11-rc3 https://lore.kernel.org/lkml/CAJZ5v0hQ3O7E3ygvFKbzjkJbx7pddyD9Qc+rE4EFYzJS12oTMg@mail.gmail.com/)
>   * Patch for Meteor Lake: https://lore.kernel.org/lkml/CAJZ5v0i8RnWs9OgxxqVOCG4RHhN7X3ELcruPoGrS1iP8hNSGsw@mail.gmail.com/
>   * Patch for Arrow Lake: https://lore.kernel.org/lkml/CAJZ5v0g+R6JK=T9EjmweRPKmcorw93JmPM4u-Z=pywv6kH7zUA@mail.gmail.com/ (both merged in 6.11-rc1 https://lore.kernel.org/lkml/CAJZ5v0h7OHR7ipj6ubgTLEz238JZkTshh93CYe_gDx5j+zBwfA@mail.gmail.com/)
> 
> 
> Pedro Henrique Kopper (1):
>    cpufreq: intel_pstate: Update Balance performance EPP for Emerald
>      Rapids
> 
> Srinivas Pandruvada (4):
>    cpufreq: intel_pstate: Allow model specific EPPs
>    cpufreq: intel_pstate: Update default EPPs for Meteor Lake
>    cpufreq: intel_pstate: Update Meteor Lake EPPs
>    cpufreq: intel_pstate: Use Meteor Lake EPPs for Arrow Lake
> 
>   drivers/cpufreq/intel_pstate.c | 46 +++++++++++++++++++++++++++++-----
>   1 file changed, 40 insertions(+), 6 deletions(-)
> 


-- 
--
Thibault



More information about the kernel-team mailing list