User Controlled Power Management

User Controlled Power Management

Hi everyone,

I need to perform fine-precision performance measurements on MIC and experience considerable variance due to DVFS.

I was tracking the frequency and voltage using 'micinfo' tool with 1 second resolution. I did 2 experiment executions of exactly the same test and the results are attached. The  first experiment was run just after resetting the coprocessor, and the initial frequency was 1.04 GHz. During the 1st experiment execution the frequency dropped from 1.04GHz to 0.84GHz. With the drop of frequency the clocksource changed from 'tsc' to less precise 'micetc'. During the 2nd experiment execution the frequency was 0.84GHz during the whole experiment, but the voltage changed as it is depicted on the chart. So can anybody clarify the following questions:

1. During changes of the voltage the frequency stays the same (except one discrete drop from 1.04GHz to 0.84GHz due to PROCHOT interrupt) according to the output of micinfo. But as far as I see in Datasheet (June, 2013) Figure 5-8 the frequency should drop together with voltage except Low Frequency Mode zone. So can I conclude from voltage change that coprocessor swithces to different P-states as voltage changes? And micinfo doesn't show exact value of the frequency, does it?

2. Is there a way to force the coprocessor to operate in the lowest possible voltage and frequency state Pn or any other user defined  (say 960000uV)?

3. As I can see in System Software Developer Guide (June, 2013) Figure 3-1 there is User Mode PM console app to contol Power Management. Does anybody know the details for that app?

Thank you!

Below is dmesg output, Power Management Configuration and micinfo output.

dmesg output (during frequency drop from 1.04GHz to 0.84GHz):

[  458.685498] PROCHOT interrupt
[  459.180135] Clocksource tsc unstable (delta = -122042579 ns)
[  459.180719] Switching to clocksource micetc
[  516.684694] RAS.get_freq: core not running at expected frequency


[  862.973207] PROCHOT interrupt
[  862.982334] Setting maxfreq to 800000
[  862.982355] Failed to update cpufreq policy
[  863.132325] Setting maxfreq to 0
[  863.132340] Failed to update cpufreq policy


Power Management was configured that way:
PowerManagement "cpufreq_off;corec6_off;pc3_off;pc6_off"

Here is the configuration of MIC I have:
        Flash Version          :
        SMC Firmware Version     : 1.15.4830
        SMC Boot Loader Version     : 1.8.4326
        uOS Version          :
        Device Serial Number      : ADKC25101431

        Vendor ID          : 0x8086
        Device ID          : 0x2250
        Subsystem ID          : 0x2500
        Coprocessor Stepping ID     : 3
        PCIe Width          : x16
        PCIe Speed          : 5 GT/s
        PCIe Max payload size     : 256 bytes
        PCIe Max read req size     : 512 bytes
        Coprocessor Model     : 0x01
        Coprocessor Model Ext     : 0x00
        Coprocessor Type     : 0x00
        Coprocessor Family     : 0x0b
        Coprocessor Family Ext     : 0x00
        Coprocessor Stepping      : B1
        Board SKU          : B1PRQ-5110P/5120D
        ECC Mode          : Enabled
        SMC HW Revision      : Product 225W Passive CS

Downloadapplication/pdf DVFS General Problem.pdf65.29 KB
5 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.


Excellent questions. I am consulting the experts here at Intel.

Note: As it is the holiday season, the experts may not be available until the new year. I will do my best to get you an answer before then.


Taylor, I have read a number of your posts at Intel blog on power management and they were very helpful. Thank you for that and for your reply!

Thanks. Your question is great because I am starting a new blog series on measuring power come the new year.


After updating MPSS version to the following one:

        Flash Version          :
        SMC Firmware Version     : 1.16.5078
        SMC Boot Loader Version     : 1.8.4326
        uOS Version          :
        Device Serial Number      : ADKC25101431

the voltage variabilty on the same test became insignificant +- 0.001V.

Leave a Comment

Please sign in to add a comment. Not a member? Join today