Re: [SOLVED] Radeon 6800 XT: 100% GPU core usage & 74 Watts when idle

2021-08-01 Thread The Wanderer
On 2021-07-31 at 17:20, piorunz wrote:

> Debian 11 patched kernel is out!
> That will fix everyone who have been affected and they didn't switched 
> to older kernel or compiled their own.
> 
> 
> linux-signed-amd64 (5.10.46+3) unstable; urgency=medium

>* Revert "drm/amdgpu/gfx9: fix the doorbell missing when in CGPG issue."
>  (Closes: #990312)
>* Revert "drm/amdgpu/gfx10: enlarge CP_MEC_DOORBELL_RANGE_UPPER to 
> cover full
>  doorbell." (Closes: #990312)

I was actually watching the unblock-request bug for that, but apparently
it doesn't get updated when the package migrates.

I'm in the updated kernel now, and I confirm the fix. gpu_busy_percent
is now 0 rather than 99, and the output of 'sensors' for amdgpu-pci-0d00
now reports a fan speed of 815 RPM (down from 1478), temperature
readings of 55, 56, and 68 degrees Celsius (down from 66, 72, and 78),
and a power reading of 40 W (down from 94).

I wasn't hearing/noticing meaningful levels of noise from the fan(s)
before, but it does sound just a bit quieter now - and of course less
power draw, less heat generation, and less energy consumption are all
good things.

-- 
   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



signature.asc
Description: OpenPGP digital signature


Re: [SOLVED] Radeon 6800 XT: 100% GPU core usage & 74 Watts when idle

2021-07-31 Thread piorunz

Debian 11 patched kernel is out!
That will fix everyone who have been affected and they didn't switched 
to older kernel or compiled their own.



linux-signed-amd64 (5.10.46+3) unstable; urgency=medium

  * Sign kernel from linux 5.10.46-3

  * [armhf] Add mdio-aspeed to nic-modules.
Thanks to Joel Stanley  (Closes: #991262)
  * Revert "drm/amdgpu/gfx9: fix the doorbell missing when in CGPG issue."
(Closes: #990312)
  * Revert "drm/amdgpu/gfx10: enlarge CP_MEC_DOORBELL_RANGE_UPPER to 
cover full

doorbell." (Closes: #990312)
  * Input: joydev - prevent use of not validated data in JSIOCSBTNMAP ioctl
(CVE-2021-3612)
  * sctp: validate from_addr_param return (CVE-2021-3655)
  * sctp: add size validation when walking chunks (CVE-2021-3655)
  * [s390x] sclp_vt220: fix console name to match device (Closes: #961056)
  * block: return the correct bvec when checking for gaps
  * sctp: fix return value check in __sctp_rcv_asconf_lookup

 -- Salvatore Bonaccorso   Wed, 28 Jul 2021 07:55:40 
+0200


On 24/07/2021 04:32, piorunz wrote:

On 23/07/2021 23:44, The Wanderer wrote:

https://gitlab.freedesktop.org/drm/amd/-/issues/1632  as being an
upstream(?) bug report which covers it. Not sure how directly applicable
those are, but just in case...


That escalated very quickly :D This bug is responsible for this. And 
patch works.


I compiled my own 5.10 kernel using Debian source, and I patched that 
one line in drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c file. Now I rebooted 
to my custom 5.10 kernel and it works :)


Let's see - before:

$ sensors
(...)
amdgpu-pci-0900
(...)
power1:   74.00 W  (cap = 272.00 W)

gpu-mon (from rickslab-gpu-utils)
┌─┬┐
│Card #   │card0   │
├─┼┤
│Model    │ Navi 21 RX │
│GPU Load %   │99  │
│Mem Load %   │0   │
│VRAM Usage % │4.839   │
│GTT Usage %  │0.802   │
│Power (W)    │75.0    │
│Power Cap (W)│272.0   │
│Energy (kWh) │0.0 │
│T (C)    │53.0    │
│VddGFX (mV)  │1143    │
│Fan Spd (%)  │0   │
│Sclk (MHz)   │2470    │
│Sclk Pstate  │1   │
│Mclk (MHz)   │1000    │
│Mclk Pstate  │3   │
│Perf Mode    │0-BOOTUP_DEFAULT    │
└─┴┘

And after:

$ sensors
(...)
amdgpu-pci-0900
(...)
power1:   34.00 W  (cap = 272.00 W)

┌─┬┐
│Card #   │card0   │
├─┼┤
│Model    │ Navi 21 RX │
│GPU Load %   │0   │
│Mem Load %   │0   │
│VRAM Usage % │6.109   │
│GTT Usage %  │1.147   │
│Power (W)    │35.0    │
│Power Cap (W)│272.0   │
│Energy (kWh) │0.004   │
│T (C)    │54.0    │
│VddGFX (mV)  │856 │
│Fan Spd (%)  │0   │
│Sclk (MHz)   │500 │
│Sclk Pstate  │0   │
│Mclk (MHz)   │1000    │
│Mclk Pstate  │3   │
│Perf Mode    │0-BOOTUP_DEFAULT    │
└─┴┘

0% GPU core use, 0% memory use, core clock is 500 MHz instead of maxed 
out at 2470 Mhz. Success!

Power usage is on par with Windows.
However, I still have 34W even if I disconnect second monitor. On 
Windows, with one monitor only power goes down to 10W and core is 10 
MHz. But my default setup it 2 monitors so I don't worry too much anyway :)





--

With kindest regards, piorunz.

⢀⣴⠾⠻⢶⣦⠀
⣾⠁⢠⠒⠀⣿⡁ Debian - The universal operating system
⢿⡄⠘⠷⠚⠋⠀ https://www.debian.org
⠈⠳⣄


[SOLVED] Radeon 6800 XT: 100% GPU core usage & 74 Watts when idle

2021-07-23 Thread piorunz

On 23/07/2021 23:44, The Wanderer wrote:

https://gitlab.freedesktop.org/drm/amd/-/issues/1632  as being an
upstream(?) bug report which covers it. Not sure how directly applicable
those are, but just in case...


That escalated very quickly :D This bug is responsible for this. And 
patch works.


I compiled my own 5.10 kernel using Debian source, and I patched that 
one line in drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c file. Now I rebooted 
to my custom 5.10 kernel and it works :)


Let's see - before:

$ sensors
(...)
amdgpu-pci-0900
(...)
power1:   74.00 W  (cap = 272.00 W)

gpu-mon (from rickslab-gpu-utils)
┌─┬┐
│Card #   │card0   │
├─┼┤
│Model│ Navi 21 RX │
│GPU Load %   │99  │
│Mem Load %   │0   │
│VRAM Usage % │4.839   │
│GTT Usage %  │0.802   │
│Power (W)│75.0│
│Power Cap (W)│272.0   │
│Energy (kWh) │0.0 │
│T (C)│53.0│
│VddGFX (mV)  │1143│
│Fan Spd (%)  │0   │
│Sclk (MHz)   │2470│
│Sclk Pstate  │1   │
│Mclk (MHz)   │1000│
│Mclk Pstate  │3   │
│Perf Mode│0-BOOTUP_DEFAULT│
└─┴┘

And after:

$ sensors
(...)
amdgpu-pci-0900
(...)
power1:   34.00 W  (cap = 272.00 W)

┌─┬┐
│Card #   │card0   │
├─┼┤
│Model│ Navi 21 RX │
│GPU Load %   │0   │
│Mem Load %   │0   │
│VRAM Usage % │6.109   │
│GTT Usage %  │1.147   │
│Power (W)│35.0│
│Power Cap (W)│272.0   │
│Energy (kWh) │0.004   │
│T (C)│54.0│
│VddGFX (mV)  │856 │
│Fan Spd (%)  │0   │
│Sclk (MHz)   │500 │
│Sclk Pstate  │0   │
│Mclk (MHz)   │1000│
│Mclk Pstate  │3   │
│Perf Mode│0-BOOTUP_DEFAULT│
└─┴┘

0% GPU core use, 0% memory use, core clock is 500 MHz instead of maxed 
out at 2470 Mhz. Success!

Power usage is on par with Windows.
However, I still have 34W even if I disconnect second monitor. On 
Windows, with one monitor only power goes down to 10W and core is 10 
MHz. But my default setup it 2 monitors so I don't worry too much anyway :)


--

With kindest regards, piorunz.

⢀⣴⠾⠻⢶⣦⠀
⣾⠁⢠⠒⠀⣿⡁ Debian - The universal operating system
⢿⡄⠘⠷⠚⠋⠀ https://www.debian.org
⠈⠳⣄