Re: [PATCH 9/9] drm/i915/gt: Restore aggressive post-boost downclocking

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Quoting Chris Wilson (2020-04-27 09:54:08)
> We reduced the clocks slowly after a boost event based on the
> observation that the smoothness of animations suffered. However, since
> reducing the evalution intervals, we should be able to respond to the
> rapidly fluctuating workload of a simple desktop animation and so
> restore the more aggressive downclocking.
> 
> References: 2a8862d2f3da ("drm/i915: Reduce the RPS shock")
> Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>

*** ANECDOTAL ***

In measurements on Icelake, the chrome power test [frankly a horrible
test that is mostly idle, and has no feedback on UX quality metrics]:

before:
7214.52,Joules,power/energy-pkg/,2302922017998,100.00,,
2927.24,Joules,power/energy-cores/,2302922022934,100.00,,
840.42,Joules,power/energy-gpu/,2302922024419,100.00,,
166620,M,i915/actual-frequency/,2302922026977,100.00,,
1905549106177,ns,i915/rc6-residency/,2302922028812,100.00,,
290532899942,ns,i915/rcs0-busy/,2302922032428,100.00,,
0,ns,i915/bcs0-busy/,2302922034987,100.00,,
0,ns,i915/vcs0-busy/,2302922034237,100.00,,

after:
6713.43,Joules,power/energy-pkg/,2228832095923,100.00,,
2802.07,Joules,power/energy-cores/,2228832104461,100.00,,
587.04,Joules,power/energy-gpu/,2228832106940,100.00,,
132124,M,i915/actual-frequency/,2228832095439,100.00,,
1957236452947,ns,i915/rc6-residency/,2228832089455,100.00,,
265365231893,ns,i915/rcs0-busy/,2228832089043,100.00,,
0,ns,i915/bcs0-busy/,2228832085764,100.00,,
0,ns,i915/vcs0-busy/,2228832084838,100.00,,

Some video playback (which is mostly pushing GL textures through to the
compositor, nothing uses libva):

before:
9512.58,Joules,power/energy-pkg/,995287278884,100.00,,
2631.12,Joules,power/energy-cores/,995287289129,100.00,,
3661.46,Joules,power/energy-gpu/,995287294283,100.00,,
715924,M,i915/actual-frequency/,995287298748,100.00,,
389009635708,ns,i915/rc6-residency/,995287303131,100.00,,
409781702935,ns,i915/rcs0-busy/,995287305584,100.00,,
0,ns,i915/bcs0-busy/,995287310428,100.00,,
0,ns,i915/vcs0-busy/,995287314214,100.00,,

after:
8013.46,Joules,power/energy-pkg/,994173360392,100.00,,
2077.22,Joules,power/energy-cores/,994173366417,100.00,,
2518.69,Joules,power/energy-gpu/,994173370072,100.00,,
464025,M,i915/actual-frequency/,994173363951,100.00,,
351530351036,ns,i915/rc6-residency/,994173363398,100.00,,
632114323426,ns,i915/rcs0-busy/,994173364895,100.00,,
0,ns,i915/bcs0-busy/,994173367723,100.00,,
0,ns,i915/vcs0-busy/,994173369162,100.00,,

I spot checked a few games during scenes where the GPU was not capped
out, and it was drawing about ~100MHz less for ~1W less.

*** ANECDOTAL ***

The other data point is a happy user in #1698 who was wise enough to
demand both smooth UX and low power.

We do have a huge issue in that we have no insight into P&P in CI.
Not even a single machine running a desktop config and playing back a
move and reporting power usage vs dropped frames. Nor even running the
battery tests to check that we do suspend in S3.

I spent some time trying to find some benchmarks we could use to measure
jank and power used and found nothing useful. The closest would be
wrapping rapl around gnome-shell-perf-tool, but that is still a long way
from capturing enough use cases.
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx



[Index of Archives]     [AMD Graphics]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux