APPLIED: [SRU][N][PATCH v3 0/6] Update default Intel EPP Values

Stefan Bader stefan.bader at canonical.com
Fri Sep 27 12:47:38 UTC 2024


On 27.09.24 11:10, Stefan Bader wrote:
> On 11.09.24 05:15, 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 values for MTL and ARL 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
>>
>> Tony Luck (1):
>>    cpufreq: intel_pstate: Switch to new Intel CPU model defines
>>
>>   drivers/cpufreq/intel_pstate.c | 128 ++++++++++++++++++++-------------
>>   1 file changed, 80 insertions(+), 48 deletions(-)
>>
> 
> Applied to noble:linux/master-next. Thanks.
> 
> -Stefan
> 
> 
This set of patches broke building  the kernel. I was almost decided to 
revert the whole batch but was able to fix things by cherry picking:

commit f055b6260eb3ef20a6e310d1e555a5d5a0a28ca0
     x86/cpu/vfm: Update arch/x86/include/asm/intel-family.h

     New CPU #defines encode vendor and family as well as model.

     Update the example usage comment in arch/x86/kernel/cpu/match.c

     Signed-off-by: Tony Luck <tony.luck at intel.com>
     Signed-off-by: Borislav Petkov (AMD) <bp at alien8.de>
     Reviewed-by: Thomas Gleixner <tglx at linutronix.de>
     Link: 
https://lore.kernel.org/r/20240416211941.9369-4-tony.luck@intel.com

Please test compile next time before submitting.
Thanks

- Stefan

-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_0xE8675DEECBEECEA3.asc
Type: application/pgp-keys
Size: 48643 bytes
Desc: OpenPGP public key
URL: <https://lists.ubuntu.com/archives/kernel-team/attachments/20240927/1bd19f22/attachment-0001.key>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL: <https://lists.ubuntu.com/archives/kernel-team/attachments/20240927/1bd19f22/attachment-0001.sig>


More information about the kernel-team mailing list