1.Environment Kernel: (drm-intel-next-queued)de67cba65944f26c0f147035bd62e30c5f456b9 enable RC6 disable RC6 IVB Average watt Watt-hour Average watt Watt-hour openarena 15.3 1.3 22.2 1.9 2.Environment Kernel: 3.8-rc5 (master)949db153b6466c6f7cad5a427ecea94985927311 Libdrm: (master)libdrm-2.4.41-4-g9dc0072e8d5a47fb0b2ada50fb897cb2c740bb40 Mesa: (9.0)1e8de8437a8a11f95eb745b92ed3db749b8c763c Xserver: (server-1.13-branch)xorg-server-1.13.2 Xf86_video_intel: (master)2.20.19-59-gc2d06c407e1c2cbbf3f7f6c4989710a799cd43d0 Cairo: (master)2560c0b6577a6380ef175cf18bb804913784632c Libva: (staging)21649988d6b532cc96f633db017d1e4369f640e9 Libva_intel_driver: (staging)b7cb38772e6f73d3c1b3465e0bc6c0009c2f5634 enable RC6 disable RC6 IVB Average watt Watt-hour Average watt Watt-hour 50.5 17.8 50.3 17.7 with the latest kernel(2), RC6 didn't save power on IVB and HSW, but with the older kernel(1), RC6 saved power.
Two things: - fps numbers for this benchmark for the different kernels. - Can you please bisect this issue?
Ouping is on Holidays until 2013/2/21, For any urgent, call 186 1685 2390. Happy Chinese Spring Festival, May you happy now and always.
Sounds like we simply not entering rc6. If you compare /sys/class/drm/card0/power/rc6_residency_ms before and after the test, do we see the GPU go into rc6 ever? Can you also please cross check with drm-intel-nightly?
kernel by default avg watt watt-hour game 63.4 11.9 cat /sys/class/drm/card0/power/rc6_residency_ms 1983261 disable RC6 avg watt watt-hour game 63.6 11.9 cat /sys/class/drm/card0/power/rc6_residency_ms 0 GPU go into rc6, but it didn't save power for game workload. when running game, we didn't set vblank_mode=0, so FPS=60 (In reply to comment #3) > Sounds like we simply not entering rc6. If you compare > /sys/class/drm/card0/power/rc6_residency_ms before and after the test, do we > see the GPU go into rc6 ever? Can you also please cross check with > drm-intel-nightly?
We need the numbers before and after the test run, to see how often it hits rc6 during the run. (Also if you can grab those values for the working kernel so that we can compare) That it enters rc6 at all suggests that it is more likely either the rps tuning values or yet-another-side-effect of the forcewake dance. Are you sure openarena is refresh-rate limited? Last time I tried, the openarena benchmark itself was disabling vsync. Presumably you are using UXA, so scanline waits aren't implemented (which would keep the GPU awake as well) and you've made sure the fb size is the same size as the mode. (It is not normally if openarena changes the resolution itself via SetVidMode.) The ultimate question is whether you are getting pageflips at all. Also difficult to guess without the avg/max fps @ resolution.
I am not sure how to check if openarena is refresh-rate limited, as Jesse suggested, running power workload should not set vblank_mode=0. the avg fps @ 1366x768 57.0 fps cat /sys/class/drm/card0/power/rc6_residency_ms cat /sys/class/drm/card0/power/rc6p_residency_ms rc6_residency_ms increase from 2132 to 2617 rc6p_residency_ms increase from 714842 to 770503 (In reply to comment #5) > We need the numbers before and after the test run, to see how often it hits > rc6 during the run. (Also if you can grab those values for the working > kernel so that we can compare) That it enters rc6 at all suggests that it > is more likely either the rps tuning values or yet-another-side-effect of > the forcewake dance. Are you sure openarena is refresh-rate limited? Last > time I tried, the openarena benchmark itself was disabling vsync. Presumably > you are using UXA, so scanline waits aren't implemented (which would keep > the GPU awake as well) and you've made sure the fb size is the same size as > the mode. (It is not normally if openarena changes the resolution itself via > SetVidMode.) The ultimate question is whether you are getting pageflips at > all. Also difficult to guess without the avg/max fps @ resolution.
Explain exactly the test you are performing and its outcomes. One obvious answer is that the game is now performing better and utilizing more power in the process. That it was able to hit rc6 before would suggest it had long periods of underutilization and that now that it consumes 3x the power with rc6=0 the behaviour of the game is vastly different and not comparable.
Daniel, After bisecting, I can't find the good point. whatever on Desktop or laptop, with the drm-intel-nightly and drm-intel-fixes branch kernel, Game worklaod didn't save power compared with disabling RC6. as Jesse suggested, running power workload should not set vblank_mode=0. openarena the avg fps @ 1366x768 57.0 fps (In reply to comment #1) > Two things: - fps numbers for this benchmark for the different kernels. - > Can you please bisect this issue?
On the other hand if the game is running at less than 60fps, we should be keeping the GPU busy *all the time* and rc6 should not be coming into effect. 15/22W -> 50/50W is a massive leap in power consumption for supposedly the same test. I think you have a major issue elsewhere or the tests are not comparable.
If we can't explain why this test consumes 40W more power between de67cba65944f26c0f147035bd62e30c5f456b9 and 949db153b6466c6f7cad5a427ecea94985927311 there's little point in proceeding.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.