Linux 4.4 DRM Pull Has Raspberry Pi Driver, AMDGPU Improvements

[phoronix]

Phoronix: Linux 4.4 DRM Pull Has Raspberry Pi Driver, AMDGPU Improvements

David Airlie sent in the Direct Rendering Manager (DRM) subsystem update today for the Linux 4.4 merge window...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-4.4-DRM-Pull

[imirkin]

Michael,

You may be interested in re-benchmarking some NVIDIA Kepler GDDR5 boards with nouveau with the fixes in this pull request. While there are some additional fixes still in-flight for some boards (generally to do with voltage control failures), this should help post-reclock stability on a bunch of them. Still not reliable to be used frequently though, there can still be bad interactions with the display unit.

Also if you have one of the boards with voltage control fail, try stepping through the diff power levels -- core might be able to reclock on one level but not another. e.g. the level 3 clock speed might require higher voltage (which might fail), so when changing to it, we'd end up keeping clocks at whatever they were. If the source is level 1 you get lower speeds than level 2. Hope that made sense.

[Adarion]

I wonder if the new kernel will make it before Christmas. That's probably a bit optimistic, though.
Oh well, seems no adv. PM / reclocking yet for the most recent AMD GPUs. I guess then it won't be before 4.5. But then, PM is a complex thing that touches a lot of others. So I'll be patient. And I don't even own a chip of the affected series, yet.


airlied "I've had to encourage Intel to try harder, which seems to be happening now."
Heh.

[Michael]

Michael,

You may be interested in re-benchmarking some NVIDIA Kepler GDDR5 boards with nouveau with the fixes in this pull request. While there are some additional fixes still in-flight for some boards (generally to do with voltage control failures), this should help post-reclock stability on a bunch of them. Still not reliable to be used frequently though, there can still be bad interactions with the display unit.

Also if you have one of the boards with voltage control fail, try stepping through the diff power levels -- core might be able to reclock on one level but not another. e.g. the level 3 clock speed might require higher voltage (which might fail), so when changing to it, we'd end up keeping clocks at whatever they were. If the source is level 1 you get lower speeds than level 2. Hope that made sense.

Yep, intend to give it a go once RC1 is out. Thanks.

[xxmitsu]

Seems like powerplay will come into play

http://cgit.freedesktop.org/~agd5f/linux/log/?h=amdgpu-powerplay

[dungeon]

Seems like powerplay will come into play

http://cgit.freedesktop.org/~agd5f/l...dgpu-powerplay

That is extraordinary news for Michael... i mean for Tonga/Fiji users.