Some AMD GPUs Affected By A Nasty Power Regression That Snuck Into Linux 4.18 Stable

[phoronix]

Phoronix: Some AMD GPUs Affected By A Nasty Power Regression That Snuck Into Linux 4.18 Stable

A Phoronix reader emailed in that since the recent Linux 4.18.10 stable kernel the power usage on his system has increased by around 50 Watts while idling... Not the overall AC system power draw being 50 Watts, but an increase of roughly that amount on the latest 4.18 stable point releases up to this point. I've now been able to reproduce as well as bisect the cause...

http://www.phoronix.com/scan.php?page=news_item&px=AMDGPU-4.18-Power-Draw

[shmerl]

4.18 is quite broken for amdgpu/Vega anyway. The monitor can't even wake up from sleep properly (DisplayPort). A very major regression. It's only fixed in 4.19 and later.

[microcode]

I also somewhat wonder why it seems nobody backports the fixes for these serious regressions.

[R41N3R]

4.18 is quite broken for amdgpu/Vega anyway. The monitor can't even wake up from sleep properly (DisplayPort). A very major regression. It's only fixed in 4.19 and later.

I do have this issue as well since a long time on Plasma-Wayland, had to disable powersaving mode for the monitor. This thing crashes my system and I have to reboot via SSH or reset :-(

[R41N3R]

I measure the power of my main system as well and in the last weeks I wondered already if there is something wrong... but I had no time to investigate.

[shmerl]

I do have this issue as well since a long time on Plasma-Wayland, had to disable powersaving mode for the monitor. This thing crashes my system and I have to reboot via SSH or reset :-(

Why don't AMD backport such fixes to stable kernels? It sounds like an important thing to do.

[shmerl]

Does this power issue happen in 4.19 release candidates?

[Marc Driftmeyer]

4.18 is quite broken for amdgpu/Vega anyway. The monitor can't even wake up from sleep properly (DisplayPort). A very major regression. It's only fixed in 4.19 and later.

It's definitely a big pile of crap on Debian.

[dwagner]

I mentioned a similar observation in https://www.phoronix.com/forums/foru...22#post1034122 and https://www.phoronix.com/forums/foru...06#post1034606 - but given the way more serious crash bugs in amdgpu, I did not bother to open a bug report of its own on this relatively minor problem.

BTW, one developer wrote in an (otherwise unrelated) bug-report that high memory/shader clocks would be required for 4k 60Hz modes - but (a) when manually forcing the GPU to the lowest frequencies for memory/shaders this does by no means cause issues (graphics operations just go slower, but no visible distortion or such), and (b) it absolutely makes no sense that default clock frequencies jump from "minimal" to "maxed out" at the transition from 4k 50Hz to 4k 60Hz - there just isn't that much of a difference between those modes.

[dungeon]

109 W idle, he says All affected people should count themselfs up, to count time until this will be fixed . Then to file bug with a data and to ask AMD to pay for that electricity

Linux kernel does not need CoC, just agreement how whenever someone introduce regressions for a longer time his company needs to pay for it Nope, any reagression - pay now here