Initial Nouveau DRM Changes Queued For Linux 4.10

[phoronix]

Phoronix: Initial Nouveau DRM Changes Queued For Linux 4.10

With Linux 4.9 there is unfortunately no new feature code for Nouveau due to being late to merge the new functionality, but that work that missed 4.9 is now staged in DRM-Next for merging to mainline when the Linux 4.10 merge window rolls along...

http://www.phoronix.com/scan.php?page=news_item&px=Nouveau-DRM-Next-4.10-1

[imirkin]

Michael, just FYI - this collection of changes should make reclocking actually work on your Kepler (and Maxwell1, i.e. GTX 750) boards.

[kmare]

imirkin that sounds awesome! is it supposed to be dynamic as well or would I have to set it manually depending on the performance I need? (nvidia GTX 650 Ti here)

[imirkin]

imirkin that sounds awesome! is it supposed to be dynamic as well or would I have to set it manually depending on the performance I need? (nvidia GTX 650 Ti here)

Nope, statically set. I forget if you can adjust cstates with those changes, or only pstates. (A pstate switch will select the highest cstate available for that pstate by default.) pstate = "power" state, which generally maps to a different vram clock frequency, cstate = "core" state, i.e. speed of the shader cores.

And do note that there's a small chance that your box will hang when doing the reclock if you're driving monitors off of the card (we don't fully understand the line buffering stuff, I believe, and VRAM goes away and the display hw gets upset). Very small, but with automatic reclocking it adds up

[kmare]

I see, that's still great news. Thank you all for your hard and invaluable work! Can't wait to try it out when a first release of the kernel comes out.

[karolherbst]

nope, no cstate selection support, it's also pretty pointless with those patches. I have some prototypes for dynamic reclocking which I will clean up and try to get merge soon, but not before next year.

[karolherbst]

Michael also with those patches, there is the config=NvBoost=2 thing needed to go to max boost, but yeah, chances are, that the gpu will draw more power than actually allowed and I know benchmarks where this actually happens.

[kernelOfTruth]

nope, no cstate selection support, it's also pretty pointless with those patches. I have some prototypes for dynamic reclocking which I will clean up and try to get merge soon, but not before next year.

There's still some issues with reclocking and the core frequency staying the same:

https://bbs.archlinux.org/viewtopic.php?id=210043

I also got these kind of issues with my GTX760 Phantom

[ 2018.628337] nouveau 0000:01:00.0: clk: failed to raise voltage: -22
[ 2018.628338] nouveau 0000:01:00.0: clk: error setting pstate 2: -22
[ 2033.359278] nouveau 0000:01:00.0: clk: failed to raise voltage: -22
[ 2033.359279] nouveau 0000:01:00.0: clk: error setting pstate 3: -22


The core frequency stays the same, but the memory clock is increased


Any ideas ?

[karolherbst]

There's still some issues with reclocking and the core frequency staying the same:

https://bbs.archlinux.org/viewtopic.php?id=210043

I also got these kind of issues with my GTX760 Phantom

[ 2018.628337] nouveau 0000:01:00.0: clk: failed to raise voltage: -22
[ 2018.628338] nouveau 0000:01:00.0: clk: error setting pstate 2: -22
[ 2033.359278] nouveau 0000:01:00.0: clk: failed to raise voltage: -22
[ 2033.359279] nouveau 0000:01:00.0: clk: error setting pstate 3: -22


The core frequency stays the same, but the memory clock is increased


Any ideas ?

full output of dmesg and your vbios.rom file from /sys/kernel/debug/dri/0/vbios.rom

those issues should be fixed with 4.10, but maybe there is something funky with yours.