Experimental Nouveau DRM Branch Yields Better GDDR5 Kepler Re-Clocking

[kernelOfTruth]

@michael

there are still some issues left on kepler cards:

1. some cards get a wrong voltage sets at higher clock, so these cards will still crash at 0f pstate
2. the change to 0f is not rock solid yet either, so sometimes there are corruption on the screen, but changing back to 07 and 0f again may fix that.

Just two pointers while doing your tests

Well, I hope I don't run into these kind of issues with your branches (crashes, hardlocks),

since I had enough of that in the past XD

I'm curious how it'll stack against the latest nvidia drivers at Unity Valley Benchmark.

Does it even run on nouveau ?

Thanks for your (and the other devs) great work !

[karolherbst]

Well, I hope I don't run into these kind of issues with your branches (crashes, hardlocks),

since I had enough of that in the past XD

I'm curious how it'll stack against the latest nvidia drivers at Unity Valley Benchmark.

Does it even run on nouveau ?

Thanks for your (and the other devs) great work !

I made some performance tests and was around 70% blob performance.

[karolherbst]

also before complaining that it doesn't work: double check that you are using the gddr5 branch, not master

[kernelOfTruth]

also before complaining that it doesn't work: double check that you are using the gddr5 branch, not master

Roger that, that branch doesn't seem to include your work

master_karol however should be fine, no ?

I'm fed up with the install, then copy-over, yadda yadda yadda approach
(also it's tricky with genkernel and the initramfs rebuilding)

Copied it over to the kernel directory and replaced the in-tree nouveau source

let's see how it goes

edit:

you're using Chromium, btw ? does it work without problems - especially bookmark handling ?

There seems to be a hardlock trigger when using newer Chromium versions:

to pre-empt this I'm also using

http://marc.info/?l=linux-kernel&m=143825266507865&w=2

[PATCH] drm/nouveau/gem: tolerate a buffer specified multiple times https://bugs.freedesktop.org/show_bug.cgi?id=91535

[karolherbst]

master_karol however should be fine, no ?

no, gddr5

[kernelOfTruth]

hm, seems like the GTX 760 Phantom is quite a beast:

no difference,

re-clocking unfortunately still fails:

[   23.448134] nouveau 0000:01:00.0: NVIDIA GK104 (0e4070a2)
[   23.667832] nouveau 0000:01:00.0: bios: version 80.04.c4.00.82
[   23.668563] nouveau 0000:01:00.0: fb: 2048 MiB GDDR5
[   23.724806] nouveau 0000:01:00.0: DRM: VRAM: 2048 MiB
[   23.724808] nouveau 0000:01:00.0: DRM: GART: 1048576 MiB
[   23.724812] nouveau 0000:01:00.0: DRM: TMDS table version 2.0
[   23.724814] nouveau 0000:01:00.0: DRM: DCB version 4.0
[   23.724817] nouveau 0000:01:00.0: DRM: DCB outp 00: 01000f02 00020030
[   23.724819] nouveau 0000:01:00.0: DRM: DCB outp 01: 02000f00 00000000
[   23.724821] nouveau 0000:01:00.0: DRM: DCB outp 02: 08011f82 00020030
[   23.724823] nouveau 0000:01:00.0: DRM: DCB outp 03: 02822f62 0f420010
[   23.724825] nouveau 0000:01:00.0: DRM: DCB outp 05: 04833fb6 0f420010
[   23.724827] nouveau 0000:01:00.0: DRM: DCB outp 06: 04033f72 00020010
[   23.724829] nouveau 0000:01:00.0: DRM: DCB conn 00: 00001030
[   23.724830] nouveau 0000:01:00.0: DRM: DCB conn 01: 01000131
[   23.724832] nouveau 0000:01:00.0: DRM: DCB conn 02: 00010261
[   23.724833] nouveau 0000:01:00.0: DRM: DCB conn 03: 00020346
[   23.724835] nouveau 0000:01:00.0: DRM: DCB conn 04: 00000460
[   23.932689] nouveau 0000:01:00.0: DRM: MM: using COPY for buffer copies
[   24.378468] nouveau 0000:01:00.0: DRM: allocated 1920x1080 fb: 0x60000, bo ffff8807f6584000
[   24.378523] fbcon: nouveaufb (fb0) is primary device
[   24.865701] nouveau 0000:01:00.0: fb0: nouveaufb frame buffer device
[   24.865703] nouveau 0000:01:00.0: registered panic notifier
[   24.874176] [drm] Initialized nouveau 1.3.0 20120801 for 0000:01:00.0 on minor 0
[  260.684289] nouveau 0000:01:00.0: clk: failed to raise voltage: -22
[  260.684302] nouveau 0000:01:00.0: clk: error setting pstate 2: -22
[  303.807070] nouveau 0000:01:00.0: clk: failed to raise voltage: -22
[  303.807075] nouveau 0000:01:00.0: clk: error setting pstate 3: -22

so 07 and 0a seem to work, but not the other states:

cat /sys/class/drm/card0/device/pstate
07: core 405 MHz memory 648 MHz AC DC *
0a: core 405-967 MHz memory 1620 MHz
0e: core 405-1293 MHz memory 6200 MHz
0f: core 405-1293 MHz memory 6200 MHz
AC: core 405 MHz memory 648 MHz

I can switch to them but there's no FPS increase in e.g. vblank_mode=0 glxgears

[karolherbst]

:

[ 260.684289] nouveau 0000:01:00.0: clk: failed to raise voltage: -22
[ 260.684302] nouveau 0000:01:00.0: clk: error setting pstate 2: -22
[ 303.807070] nouveau 0000:01:00.0: clk: failed to raise voltage: -22
[ 303.807075] nouveau 0000:01:00.0: clk: error setting pstate 3: -22

that's because voltaging your card doesn't work yet (for whatever reasons).

Though with my patch you should be able to use the card on 0e and 0f (reclock to 0a first, then to 0e, 0f) and it should give you a perf boost from the higher memory clock

[karolherbst]

I can switch to them but there's no FPS increase in e.g. vblank_mode=0 glxgears

I would be suprised if anything would change to glxgears, you should rather do test on demanding stuff like unigine heaven

[kernelOfTruth]

I would be suprised if anything would change to glxgears, you should rather do test on demanding stuff like unigine heaven

Right, you are

Unigine Valley
2x AA, 1920x1080
Medium

nvidia 355.11
68.6
2871
min 31.6
max 112.8

nouveau 0a
16.4
687
min 7.2
max 29.7

nouveau 0f
16.6
695
min 11.0
max 29.7


The Stability is definitely an improvement

it used to hardlock pretty quickly

[   23.448134] nouveau 0000:01:00.0: NVIDIA GK104 (0e4070a2)
[   23.667832] nouveau 0000:01:00.0: bios: version 80.04.c4.00.82
[   23.668563] nouveau 0000:01:00.0: fb: 2048 MiB GDDR5
[   23.724806] nouveau 0000:01:00.0: DRM: VRAM: 2048 MiB
[   23.724808] nouveau 0000:01:00.0: DRM: GART: 1048576 MiB
[   23.724812] nouveau 0000:01:00.0: DRM: TMDS table version 2.0
[   23.724814] nouveau 0000:01:00.0: DRM: DCB version 4.0
[   23.724817] nouveau 0000:01:00.0: DRM: DCB outp 00: 01000f02 00020030
[   23.724819] nouveau 0000:01:00.0: DRM: DCB outp 01: 02000f00 00000000
[   23.724821] nouveau 0000:01:00.0: DRM: DCB outp 02: 08011f82 00020030
[   23.724823] nouveau 0000:01:00.0: DRM: DCB outp 03: 02822f62 0f420010
[   23.724825] nouveau 0000:01:00.0: DRM: DCB outp 05: 04833fb6 0f420010
[   23.724827] nouveau 0000:01:00.0: DRM: DCB outp 06: 04033f72 00020010
[   23.724829] nouveau 0000:01:00.0: DRM: DCB conn 00: 00001030
[   23.724830] nouveau 0000:01:00.0: DRM: DCB conn 01: 01000131
[   23.724832] nouveau 0000:01:00.0: DRM: DCB conn 02: 00010261
[   23.724833] nouveau 0000:01:00.0: DRM: DCB conn 03: 00020346
[   23.724835] nouveau 0000:01:00.0: DRM: DCB conn 04: 00000460
[   23.932689] nouveau 0000:01:00.0: DRM: MM: using COPY for buffer copies
[   24.378468] nouveau 0000:01:00.0: DRM: allocated 1920x1080 fb: 0x60000, bo ffff8807f6584000
[   24.378523] fbcon: nouveaufb (fb0) is primary device
[   24.865701] nouveau 0000:01:00.0: fb0: nouveaufb frame buffer device
[   24.865703] nouveau 0000:01:00.0: registered panic notifier
[   24.874176] [drm] Initialized nouveau 1.3.0 20120801 for 0000:01:00.0 on minor 0
[  260.684289] nouveau 0000:01:00.0: clk: failed to raise voltage: -22
[  260.684302] nouveau 0000:01:00.0: clk: error setting pstate 2: -22
[  303.807070] nouveau 0000:01:00.0: clk: failed to raise voltage: -22
[  303.807075] nouveau 0000:01:00.0: clk: error setting pstate 3: -22
[  924.625638] nouveau 0000:01:00.0: clk: failed to raise voltage: -22
[  924.625646] nouveau 0000:01:00.0: clk: error setting pstate 3: -22

[kernelOfTruth]

cat /sys/class/drm/card0/device/pstate

07: core 405 MHz memory 648 MHz AC DC *
0a: core 405-967 MHz memory 1620 MHz
0e: core 405-1293 MHz memory 6200 MHz
0f: core 405-1293 MHz memory 6200 MHz
AC: core 405 MHz memory 648 MHz

cat /sys/class/drm/card0/device/pstate

07: core 405 MHz memory 648 MHz
0a: core 405-967 MHz memory 1620 MHz AC DC *
0e: core 405-1293 MHz memory 6200 MHz
0f: core 405-1293 MHz memory 6200 MHz
AC: core 966 MHz memory 1620 MHz

cat /sys/class/drm/card0/device/pstate

07: core 405 MHz memory 648 MHz
0a: core 405-967 MHz memory 1620 MHz
0e: core 405-1293 MHz memory 6200 MHz AC DC *
0f: core 405-1293 MHz memory 6200 MHz
AC: core 966 MHz memory 6199 MHz

cat /sys/class/drm/card0/device/pstate

07: core 405 MHz memory 648 MHz
0a: core 405-967 MHz memory 1620 MHz
0e: core 405-1293 MHz memory 6200 MHz
0f: core 405-1293 MHz memory 6200 MHz AC DC *
AC: core 966 MHz memory 6199 MHz