amdgpu 0000:0a:00.0: GPU reset begin! after some time played

[zaggynl]

Your system information

• System information from steam (Steam -> Help -> System Information) in a gist:
  https://gist.githubusercontent.com/zaggynl/0d8906e962ff1e584fd37a0dcb541d62/raw/1a29c1cb1c501c7463cb651e553b9821b7b4b4b5/system%2520info
• Have you checked for system updates?: [Yes/No] Yes
• Are you using the latest stable video driver available for your system? [Yes/No] Yes
• Have you verified the game files?: [Yes/No] Yes

Please describe your issue in as much detail as possible:

Describe what you expected should happen and what did happen. Please link any large pastes as a Github Gist.
After some time playing screen freezes, sounds/network continue, using voicecom with friends.
syslog shows: https://gist.github.com/zaggynl/87c57f9ee0ffffbfadaa5afb7367cfb7
Using kernel 5.0.0-27-generic (Ubuntu 18.04 HWE kernel)
ACO PPA for Mesa per https://steamcommunity.com/app/221410/discussions/0/1640915206474070669/
Videocard: RX Vega 56 reference model with everything stock.
PSU: Seasonic Focus Plus 550W Platinum

Steps for reproducing this issue:

1. Play a couple rounds Dota 2 against bots
2. ???
3. GPU driver reset, have to reset PC

[kisak-valve]

Hello @zaggynl, can you reproduce the issue with RADV_PERFTEST=llvm %command% in the game's launch options?

If the game works fine with that launch option, then that hints you've encountered an issue with the experimental driver and it should reported over at https://github.com/daniel-schuermann/mesa/issues.

[zaggynl]

Can confirm this occurs with and without the RADV_PERFTEST=llvm %command% launch option.
Am adding temperature monitoring with below:

#!/bin/bash
while true
do
        file="/sys/class/drm/card0/device/hwmon/hwmon1/temp1_input"
        value=$(cat "$file")
        divide=1000
        temp=$((value / divide))
        timestamp=$(date +"%Y-%m-%d %T")
        CEL=$'\xe2\x84\x83 '
        echo $timestamp $temp$CEL >> gputemplog.txt
        echo "$timestamp $temp$CEL and logging to $PWD/gputemplog.txt"
        sleep 1
done

Edit:
Seems to hover around 75°C, I'll try forcing fanspeed 120/255 with:

echo 1  >   /sys/class/drm/card0/device/hwmon/hwmon1/pwm1_enable 
echo 120  > /sys/class/drm/card0/device/hwmon/hwmon1/pwm1

[zaggynl]

fanspeed did not make a change

will now try opengl instead of vulkan renderer

[zaggynl]

No differences between renderer.
Now trying borderless window instead of desktop-friendly fullscreen.

[zaggynl]

No differences with borderless window mode, enabled v-sync.

[zaggynl]

Still crashes with v-sync enabled, I'm not sure what to try else.
Should I provide any other logs to debug?

Trying this replay https://medium.com/layerth/benchmarking-dota-2-83c4322b12c0.
Started Dota 2 with launch options +exec_async benchmark +demo_quitafterplayback
But did not crash.
If it crashes again in one of my games I'll save the replay, maybe I can reproduce the crash that way.

Edit; some people are having luck with R600_DEBUG=nodcc %command%, will try that.

[zaggynl]

R600_DEBUG=nodcc %command% made no difference

Appear unable to download match replay, remains stuck at downloading: http://i.imgur.com/kG3zmUf.png

Tried through https://www.opendota.com/matches/5051516611, also no luck.

Can't find the match ID in: ~/.steam/steam/steamapps/common/dota 2 beta/game/dota/replays/
Manually placed replays work.

[zaggynl]

Tried using Proton 4.11-6, GPU also appears to crash a couple minutes into a botmatch, no entries left in syslog, just a line of @ symbols.

[zaggynl]

No differences between renderer.
Now trying borderless window instead of desktop-friendly fullscreen.

Reading back, seeing as my GPU also crashed during opengl it is no long relevant to this git repo.
This is likely a heat issue with the card itself, I've swapped it with a loaner Geforce GTX 1070 Ti and all issues disappeared.

Gave some thought to undervolting but apparently this proves troublesome under Linux:
Have to fool around with powertables?
https://forum.level1techs.com/t/how-to-overclock-vega-on-linux/132771/66
ROCm/ROCm#463