Announcement

Collapse
No announcement yet.

AMD Linux 3.12 Tests With The Performance Governor

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • AMD Linux 3.12 Tests With The Performance Governor

    Phoronix: AMD Linux 3.12 Tests With The Performance Governor

    Complementing the article earlier today why Radeon graphics are faster on Linux 3.12, here's some benchmarks now showing the open-source Radeon results when using the CPUfreq performance governor on Linux 3.12 with the Core i7 4770K Haswell...

    http://www.phoronix.com/vr.php?view=MTQ4NjE

  • #2
    So basically there is no need to force the performance governor with the upcoming 3.12 kernel.

    Comment


    • #3
      Great post and great job for all this benchmarkings.
      It would be cool a radeon free driver + kernel 3.12 + dpm + mesa 10.0 vs catalyst. Maybe we could be surprised about the results

      Comment


      • #4
        I would appreciate some testing on AMD CPUs as well though, just to see how general this is.

        Comment


        • #5
          Originally posted by edoantonioco View Post
          Great post and great job for all this benchmarkings.
          It would be cool a radeon free driver + kernel 3.12 + dpm + mesa 10.0 vs catalyst. Maybe we could be surprised about the results
          And catalyst should be run with performance governor, otherwise it won't be fair.

          Comment


          • #6
            Michael, what does Intel P-State look like compared to 3.11 and 3.12 ondemand?

            Comment


            • #7
              Originally posted by nslqqq View Post
              And catalyst should be run with performance governor, otherwise it won't be fair.
              Another thing, with the "performance" flag activated in /sys/class/drm/card0/device/power_dpm_state. That helps the DPM a lot, not sure if it should be necessary done for every ati card.

              Comment


              • #8
                @Michael:
                The actual power consumption would be interesting for 3.11 vs 3.12. Maybe the new defaults for the ondemand governor allow longer sleeping periods.

                Also, could you run the same set of benchmarks (3.11 ondemand, 3.12 ondemand, 3.12 performance) for an AMD APU?
                Last edited by oleid; 10-15-2013, 02:12 PM.

                Comment


                • #9
                  One interesting case seems to be missing: 3.11 with performance gouvernor.

                  Comment


                  • #10
                    Originally posted by AnonymousCoward View Post
                    One interesting case seems to be missing: 3.11 with performance gouvernor.
                    Might as well try it with 3.12 too

                    I'd like to see the comparative results of:

                    Intel platform w/ HD 6950
                    - (broken p-state, aka with cpufreq) ondemand, performance
                    - with p-state
                    - 3.11 v 3.12
                    - catalyst vs OSS
                    - requires 12 runs

                    AMD platform w/ discrete HD 6950
                    - ondemand, performance
                    - 3.11 v 3.12
                    - catalyst vs OSS
                    - requires 8 runs

                    AMD platform w/ high end APU (say an A10)
                    - ondemand, performance
                    - 3.11 v 3.12
                    - catalyst vs OSS
                    - requires 8 runs

                    Michael, I expect the results to be on my desk by 5PM

                    Comment


                    • #11
                      Oh yes, I forgot about the Iris Pro tests I'd like to see too (in comparison to maybe a 4600) ... but you can hand those ones in in the morning (just to prove that I'm not all evil)

                      Comment


                      • #12
                        Amd cpu...

                        @Michael:
                        An AMD CPU man please! Make our day more!! Don't forget us the Phenom II users! After all Phenom II was one of the most honest processor family ever built in the industry...

                        BTW I never used adblock or bull***t like that!

                        I tried to run PTS here on my opensuse 12.3 system cause I use kernel 3.12 rc4 and all latest Xorg and Mesa stuff from Pontostroy's repository but when loading the profile 17 gaming the warsow game download stuck!
                        Also opensuse does not provide curl3 anymore and this ia32 package that PTS wanted as dependencies! I found and installed the others that asked me though...
                        The PTS pankage for opensuse 12.3 is 4.2.0 latest! Why is that old? I know PTS is in version 4.8.3 now!

                        Comment


                        • #13
                          xonotic test with radeon hd 6850 with ondemand and perfomance. cpu is and old phenon @2500ghz

                          http://openbenchmarking.org/result/1...SO-MERGE669574

                          Comment


                          • #14
                            It would be interesting to see a test with cpufreq completely disabled in the kernel. Many people don't need it at all, since the CPU can do the job on its own just fine, switching between different speeds depending on current load.

                            Comment


                            • #15
                              Misleading tests if the 3.11 results were obtained with the ondemand governor.

                              The least you could do since you chose to keep the 3.11 results on the same graph would be to specify that, Michael...

                              Comment

                              Working...
                              X