No announcement yet.

Intel IvyBridge/Haswell/Broadwell/Skylake OpenGL & Vulkan Benchmarks On Linux 4.10 + Mesa 13.1

  • Filter
  • Time
  • Show
Clear All
new posts

  • #11
    I guess you can't use the intel_pstate frequency governor with the AMD chip, nevertheless I feel like pointing out that on my machine (with kernel 4.8) the "acpi-cpufreq ondemand" appears to be about 1.6x slower for pure CPU code, compared to "intel_pstate powersave".

    That's for a test that runs more than 10 secs, and remains the same if the test is run repeatedly.

    "acpi-cpufreq performance" is more than 2x faster, and just slightly slower than "intel_pstate performance", in my test.

    ("intel_pstate performance" compared to "intel_pstate powersave" is about 1.5x faster.)

    The performance modes of each also have the least variation in execution times, so to me they seem best suited for performance comparisons. However I haven't tested this as throughly as would be necessary, and it may have changed for kernel 4.10 since 4.8, and also it might be specific to my hardware.


    • #12
      Originally posted by indepe View Post
      ...and also it might be specific to my hardware.
      And I suppose it is: I used
      watch grep \"cpu MHz\" /proc/cpuinfo
      to monitor CPU frequency, and although the displayed frequency for one of the CPUs will quickly more than double when I run my test, it will have little effect on the execution time. I suppose the dynamic changes of "powersave" and "ondemand" are not getting through to the hardware correctly. However setting different fixed frequencies with "userspace" has the expected effect, for example.