On 2021-07-23 at 16:44, piorunz wrote: > Hello, > > I am using Bullseye 11 with Radeon 6800 XT and noticed higher > temperature and noise comparing to Windows, so I investigated this and > found the fault. GPU core works at 100% usage at all times, even at idle. > > $ cat /sys/class/drm/card0/device/gpu_busy_percent > 99 That's interesting. I get the same output from that command (with a 5600 XT), but I'm not noticing any meaningful noise that seems to be coming from the card. > $ sensors > (...) > amdgpu-pci-0900 > Adapter: PCI adapter > vddgfx: 1.14 V > fan1: 1098 RPM (min = 0 RPM, max = 3000 RPM) > edge: +51.0°C (crit = +100.0°C, hyst = -273.1°C) > (emerg = +105.0°C) > junction: +55.0°C (crit = +110.0°C, hyst = -273.1°C) > (emerg = +115.0°C) > mem: +56.0°C (crit = +100.0°C, hyst = -273.1°C) > (emerg = +105.0°C) > power1: 74.00 W (cap = 272.00 W) I get 1478 RPM, and temperature readings of 66, 72, 78, plus a power reading of 94 W. Those do sound higher than reasonable, but I'm disinclined to trust the RPM value at least, since that'd be nearly 25 revolutions per second and I'm nearly positive that'd be a lot more audible than what I'm hearing. And of course if that reading isn't accurate, there's less reason to trust the others. > gpu-mon (from rickslab-gpu-utils) Where does that come from? It's not in current Debian stable or testing. > This is my hardware: > $ inxi -G > Graphics: Device-1: Advanced Micro Devices [AMD/ATI] Navi 21 [Radeon RX > 6800/6800 XT / 6900 XT] driver: amdgpu v: kernel > Display: x11 server: X.Org 1.20.11 driver: loaded: > amdgpu,ati unloaded: fbdev,modesetting,radeon,vesa resolution: > 1: 1920x1200~60Hz 2: 3840x2160~60Hz > OpenGL: renderer: AMD SIENNA_CICHLID (DRM 3.40.0 > 5.10.0-8-amd64 LLVM 11.0.1) v: 4.6 Mesa 20.3.5 I get: Graphics: Device-1: AMD Navi 10 [Radeon RX 5600 OEM/5600 XT / 5700/5700 XT] driver: amdgpu v: kernel Display: server: X.Org 1.20.11 driver: loaded: amdgpu,ati unloaded: fbdev,modesetting,radeon,vesa resolution: 2560x1600~60Hz OpenGL: renderer: AMD Radeon RX 5700 XT (NAVI10 DRM 3.40.0 5.10.0-8-amd64 LLVM 12.0.1) v: 4.6 Mesa 21.1.4 Mesa and a couple of other things are from sid, courtesy of an experiment I did on the wrong path to getting Vulkan working, but otherwise this looks substantially similar. > All packages are default from Bullseye (apart from rickslab-gpu-utils > package which I installed to debug this issue), fresh install from the > scratch a week ago, all graphics and games works perfectly. It's just > that 100% GPU usage when idle. 74 Watts used when idle! I did further > testing and found out this: > > Windows 10 > one monitor: 9-11 W, GPU clock 0-10 MHz, Memory Clock 40-100 MHz > two monitors: 34 W, GPU clock 0-10 MHz, Memory Clock 2000 MHz > > Debian 11 > one monitor: 74 W, GPU clock 2475 MHz, Memory Clock 2000 MHz > two monitors: 74 W, GPU clock 2475 MHz, Memory Clock 2000 MHz Where are the Debian-side results from? > Debian results are very bad! On Windows 10, high memory clock with two > monitors it's a known issue, but on Debian also 100% GPU usage and > 2.5GHz GPU core clock is a big problem. > > Any tips appreciated. My initial guess is that this is either A: bad data / false positive (i.e., it's not really that bad under the hood), or B: a limitation of the Linux-side drivers. If you're willing to try out the official direct-from-AMD drivers (which may or may not include an associated stack, I'm not sure), and they give different results, that could be informative. I'm not willing to try those out under present circumstances, so I can't directly contribute information there. -- The Wanderer The reasonable man adapts himself to the world; the unreasonable one persists in trying to adapt the world to himself. Therefore all progress depends on the unreasonable man. -- George Bernard Shaw
Attachment:
signature.asc
Description: OpenPGP digital signature