Turbo boost when all cores are busy?

[DDavi34]

Hi all,

I see an unexpected (to me) behavior on an Intel(R) Xeon(R) CPU E5-2699 v3 @ 2.30GHz.

If I disable turbo boost and set the performance governor with intel_pstate on a recent Linux kernel, the p-state is fixed at 2.3GHz, which is what I expect.

if I enable turbo and keep all cores busy with a tight loop doing some math, then the pstate appears to be 2.8GHz (and the performance I measure corresponds); this is unexpected to me.

I thought that the maximum p-state with all cores busy would not change by enabling and disabling turbo boost.

Why would there be some all-cores-busy turbo frequencies above the nominal processor frequency?

My problem is that I need to take some measurements and I would like to configure the system to stay at the maximum p-state regardless of load.

Disabling turbo boost and setting the performance governor is the way I used to do this, but it seems that I am leaving some performance on the table that way, because I get 2.3GHz instead of the 2.8GHz that I get with turbo boost enabled.

[idata]

Hi dbb,

I have sent your question to my upper level for further details on how the processor will work with your customization. Take in mind Intel® Turbo Boost Technology is a way to automatically run the processor core faster than the marked frequency if the part is operating under power, temperature, and current specifications limits of the thermal design power (TDP). This results in increased performance of both single and multi-threaded applications. Each Core has its own maximum speed with Turbo Boost, more details on page 9.http://www.intel.com/content/dam/www/public/us/en/documents/specification-updates/xeon-e5-v3-spec-update.pdf http://www.intel.com/content/dam/www/public/us/en/documents/specification-updates/xeon-e5-v3-spec-update.pdf Regards,Mike C

[Ronny_G_Intel · Accepted solution]

HI dbb,

I really dont know if I captured your question correctly but let me see if I can help.

Your processor Intel(R) Xeon(R) CPU E5-2699 v3 @ 2.30GHz with Turbo Boost disable runs at 2.3GHz as expected.

If you enable Turbo Boost then the processors runs at 2.8Ghz. When setting up p-state at 2.3Ghz but this is being affected by Turbo Boost. Remember that P-state is both a frequency and voltage operating point. Both are scaled as the P-state increases so I would say this is the expected behavior.

Are you working on some kind of programing? If yes, perhaps our https://software.intel.com/en-us/forum https://software.intel.com/en-us/forum is a better resource for this concern.

Regards,

Ronny G

[idata]

Hi dbb,

Do you have any update in regards "rguevara" request? Mike C

[DDavi34]

Hi Mike,

Sorry, I've been super busy, haven't had time to reply.

Thanks both for your replies.

It turns out that the behavior I see is fine, my expectations were wrong.

I did not know that turbo boost still has headroom to exploit even when all cores are busy, I wasn't considering that the instruction mix will matter.

I haven't done this experiment, but I guess that if I ran an AVX-heavy workload on all cores the resulting p-state will be lower and close to the nominal one.

We can close this thread.

Thanks!

[idata]

Hi dbb,

Thank you for your feedback. Mike C