Announcement

Collapse
No announcement yet.

Nouveau Driver Power Management Against The NVIDIA Blob

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

  • Nouveau Driver Power Management Against The NVIDIA Blob

    Phoronix: Nouveau Driver Power Management Against The NVIDIA Blob

    Following last week's completion of the Radeon driver power management tests against the AMD Catalyst driver, now it is time to turn the tables on NVIDIA. In this article are some power consumption and thermal tests when comparing the latest open-source "Nouveau" driver code against NVIDIA's closed-source proprietary driver.

    http://www.phoronix.com/vr.php?view=16195

  • #2
    Just a quick bug report for your PTS Graph Generator

    It should be Less is Better and More is Better not Less are Better and More are Better

    Comment


    • #3
      Looks like nouveau power consumption is coming along nicely — or else that it's just not very significant compared to that of a whole desktop machine anyway (only reading idle and DVD-playback graphs; power usage during games is a bit hard to compare when performance is so different).

      Comment


      • #4
        Could you try to run this on a laptop? My Thinkpad T61p with the Quadro 570FX battery life was quite shorter when using nouveau instead of the binary blob. If the newer version is better I could consider switching back. BTW, you should add playing a Flash video to the test, it suck ups battery like nothing!

        Comment


        • #5
          these benchmarks have not so much sense as you check only what can do the not official driver .
          may be checking changes in the kernel to make the ati/nv card less consuming with the official driver is a better useful thing to do ;']

          Comment


          • #6
            Originally posted by Cyborg16 View Post
            Looks like nouveau power consumption is coming along nicely or else that it's just not very significant compared to that of a whole desktop machine anyway (only reading idle and DVD-playback graphs; power usage during games is a bit hard to compare when performance is so different).
            It looks that way, because Michael was careful to stash the FPS graphs on a different page. 100W may look better than 110W, but not so much when you realize you're getting 30FPS for 100W vs 130FPS for 110W. (I just made those numbers up, the article is very hard to follow.)

            Comment


            • #7
              Originally posted by bug77 View Post
              It looks that way, because Michael was careful to stash the FPS graphs on a different page. 100W may look better than 110W, but not so much when you realize you're getting 30FPS for 100W vs 130FPS for 110W. (I just made those numbers up, the article is very hard to follow.)
              That's a good idea... FPS per Watt. We all know Nouveau is still trailling the binary blob, but would be nice to see.

              Comment


              • #8
                I did not notice that nouveau had active powermangement. Usually it is possible the change the powerlevel manually using a module option. Thats also possible via sysfs when you activated the override. I never saw that it would change the performancelevel automatically. So what was most likely compared is the default boot powerlevel that is stored in the firmware. Look at dmesg to see all possible, ask in the irc://irc.freenode.net/#nouveau how to enable the switching. It will increase speed very much when you use the fastest performance mode. Most likely you can save energy when you enable a lower state. Tests without comparing the frequency of the gpu is a bit boring.

                Comment


                • #9
                  CPU usage

                  WRT the graphs on page 6 of the test results, showing more frames per second with the proprietary driver, then the comment that it is 'interesting' that the Nouveau driver uses less of the CPU, well, this is exactly what you would expect, and is not a surprise at all, because the increased frames per second is directly the result of the increased usage of the CPU by the proprietary driver.

                  Comment


                  • #10
                    Originally posted by Kano View Post
                    I did not notice that nouveau had active powermangement. Usually it is possible the change the powerlevel manually using a module option.
                    I was thinking the same thing, but then saw that voltage and freq adjustments are done in nouveau. But AFAIK, they do not switch automatically.

                    A nouveau dev has recently commented that nouveau runs the chips at 50MHz, and the blob at 600Mhz. That explains the "saving".

                    This seems a bit like running the low profile all the time with AMD cards.

                    Comment


                    • #11
                      Isn't worth much

                      I have an HP desktop with nVidia GeFoorce 7100 video. Anything such as Fedora 15 or open SUSE trying to use the Nouveau drive either out and out kills my video or eventually caused everything to lock up. It may be great for tablets and netbooks but it isn't worth a shit for desktops!

                      Comment


                      • #12
                        watts per frame in one and frequency/temperature/power in another graph would be very much appreciated

                        Comment


                        • #13
                          Originally posted by jakubo View Post
                          watts per frame in one and frequency/temperature/power in another graph would be very much appreciated
                          Just divide the number of W by the number of FPS. Of course, it would be easier if the graph were on the same page, but you can do it regardless.

                          Comment


                          • #14
                            Originally posted by Remote User View Post
                            WRT the graphs on page 6 of the test results, showing more frames per second with the proprietary driver, then the comment that it is 'interesting' that the Nouveau driver uses less of the CPU, well, this is exactly what you would expect, and is not a surprise at all, because the increased frames per second is directly the result of the increased usage of the CPU by the proprietary driver.
                            You totally stole my post. The driver is indeed the bottleneck there.

                            Comment


                            • #15
                              Originally posted by Phoronix
                              Interestingly, when using the NVIDIA driver its core temperature was higher than the Nouveau driver for the GeForce 8 series. For the GeForce 9 series, the core temperature was lower. This is quite odd that the NVIDIA driver produced a higher core temperature while being more power efficient, but perhaps there is a bug in the GeForce 8 series reverse-engineered thermal sensor code that is causing the core temperature to be lower than it actually is running.
                              Have you thought about monitoring fan speeds? The nvidia binary driver switches between different performance levels according to the gpu-load.
                              In many graphic cards bios a fan speed value is set for each performance level
                              (take a look at nvidia-settings powermizer).
                              The nouveau driver, like earlier stated, does not automatically switch between different performance levels.
                              Overall it would be much much more interesting to see, if it is possible to lower the power consumption of the nouveau driver.

                              Would be great if the article could be updated...

                              Comment

                              Working...
                              X