PC Drivers & Software

[lalala]

my RX 480 card have random issue of crash or freeze or black screen, under idle/suspend/wake/low load conditions,

this issue happen in both windows 10 and linux , and require power/reset button

issue occur with original video bios, with latest official F5 video bios, with multiple versions of windows and linux

card is gigabyte rx480 g1 gaming 4gb
cpu ryzen 5 1600
motherboard asrock b450 pro4
memory ddr4 32gb 3000
windows 10 2004 and linux 5.4

from many testings in linux, highlight the following observations, still true as of recent, Arch Linux kernel 5.4.77

my RX 480 video card bios specify idle 800mv voltage,  it is actually run 750mv at idle

linux sensors output, pc is at idle,  video bios idle value is 800mv,  it is actually reported running 750mv

issue may occur multiple times a day

amdgpu-pci-0a00
Adapter: PCI adapter
vddgfx: 750.00 mV
fan1: 802 RPM (min = 0 RPM, max = 3700 RPM)
edge: +29.0°C (crit = +94.0°C, hyst = -273.1°C)
power1: 8.20 W (cap = 130.00 W)

linux sensors output, pc is at idle, video bios modified idle value at 820mv, it is actually reported running 775mv,

no more random crash or freeze or black screen  for days/weeks/months

amdgpu-pci-0a00
Adapter: PCI adapter
vddgfx: 775.00 mV
fan1: 796 RPM (min = 0 RPM, max = 3700 RPM)
edge: +29.0°C (crit = +94.0°C, hyst = -273.1°C)
power1: 8.19 W (cap = 130.00 W)

for William and for people who experience same issue,  assume your pc run windows 10 and amd video driver,
please check idle voltages, may try configure additional 25mv or 50mv at idle,   and report result

for AMD,    please make video driver to respect video bios idle value of 800mv, do not run it at 750mv

compared rx480 and rx580 card bios,  noticed that RX580 video bios set idle value at 750mv

so i suspect AMD video driver treats both rx480 and rx580 to idle at same 750mv , which likely is contributor of  these random crash/freeze/blackout issues under idle/suspend/wake/low load conditions

many of these very similar occurrences are reported on internet
just google for keywords:  rx 480 (or polaris)  crash or freeze or black screen

[mstfbsrn980]

https://community.amd.com/t5/knowledge-base/information-required-when-posting-a-discussion/ta-p/4227...

The vast majority of GPUs remain stable in idle mode with 750mV voltage. The problem you're having is probably something different...

[lalala]

quote:  The vast majority of GPUs remain stable in idle mode with 750mV voltage.

is your statement in agreement or disagreement of this thread's claim ?
it certainly leaves grey area for interpretation

what about the discrepancy that video bios says to run its idle at 800mv,  but driver software seems to run idle at 750mv
i am pretty sure 800mv was determined and set for a reason by AMD

quote:  The problem you're having is probably something different...

you are subtly exploring other potential cause, which is valid to raise

i will try to provide additional info, and use polarisbioseditor tool to show its clock/voltage table

#1
original F5 video bios,    retested its original bios in linux 4 days agao,  archlinux 5.4 kernel

result:  during 2 days it was on original video bios, the issue happened multiple times

mhz mv
300 800
608 65282
910 65283
1077 65284
1145 65285
1191 65286
1236 65287
1290 65288
memory
300 1000
1750 1000

#2
modified video bios,   it was on this bios for about half a day

result:  the issue still occur

mhz mv
300 800
600 65282
900 65283
1000 65284
1050 65284
1100 65285
1150 65285
1200 65286
memory
300 1000
1750 1000

#3
modified video bios,   it being on this bios for about 3 days

result:  the issue no longer occur, system is stable

mhz mv
300 820
600 65282
900 65283
1000 65284
1050 65284
1100 65285
1150 65285
1200 65286
memory
300 1000
1750 1000

in modified bios,  while its clock is capped/less-clock than original,  voltage is at over-volt/about-same as official entries

between mod-1 and mod-2 , the difference is in idle volt value (800mv vs 820mv)

result is reported run time idle volt is  (750mv) vs (775mv)

impact is   random crash/freeze/blackscreen vs  a stable system

do you/anyone still want to make argument that idle 750mv is ok and issue cause is elsewhere ?

i think AMD driver run both rx580 and rx480 at idle 750mv is a contributor to this kind of random crash/freeze/blakscreen at idle/suspend/wake/low-load situations
pretty sure bios idle value 800mv was determined and set for a reason by AMD

 would like AMD team to

1. make video driver to respect what video bios says to run it at

2. if possible, expose tool so that end-users can generate signed uefi video bios with custom configurations,

third party polarisbioseditor tool is not ideal

[lalala]

in case anyone would like to see a test case that use official video bios as baseline,

test plan:    

1. use RX480 card with official video bios , to verify if the issue is still there

2. use RX480 card with official video bios modified to set idle value 820mv (all other values kept same as in official),  to verify if the issue is still there ,    polarisbioseditor.exe tool is used to modify video bios

OS is arch linux 5.4.77 kernel

test results

test case 1 ,   with official video bios,  bios idle value 800mv ,
linux sensors report it idle at 750mv , 
within 2 hours, a freeze happened, with led lights flashing on video card (same as in previous crash/freeze/blackout occurrences) , have to use reset or power button to restart computer

test case 2 ,  with modified video bios,  bios idle value 820mv (only this value is customized) ,
linux sensors report it idle at 775mv ,
about 5 hours so far,  system is stable, the issue no longer occur

for purpose of this validation test,  will update in a few more days

recap:

the issue is  RX480 randomly crash/freeze/blackout  in  idle/suspend/wake/low-load situations

with video card's official video bios,  the issue exist

with higher run idle voltage,  the issue has disappeared

video bios set idle value at 800mv,  sensors report it idle at 750mv  (the issue occur)

video bios set idle value at 820mv,  sensors report it idle at 775mv  (the issue disappear)

if any question, please feel free to share

[lalala]

in case anyone would like to see a test case that use official video bios as baseline,

test plan:

1. use RX480 card with official video bios , to verify if the issue is still there

2. use RX480 card with official video bios modified to set idle value 820mv (all other values kept same as in official), to verify if the issue is still there , polarisbioseditor.exe tool is used to modify video bios

OS is arch linux 5.4.77 kernel

test results

test case 1 , with official video bios, bios idle value 800mv ,
linux sensors report it idle at 750mv ,
within 2 hours, a freeze happened, with led lights flashing on video card (same as in previous crash/freeze/blackout occurrences) , have to use reset or power button to restart computer

test case 2 , with modified video bios, bios idle value 820mv (only this value is customized) ,
linux sensors report it idle at 775mv ,
so far, about 2 days (total power on more than 10 hours),  no more random crash/freeze/black-screen

recap:

the issue is RX480 randomly crash/freeze/blackout in idle/suspend/wake/low-load situations

with video card's official video bios, the issue exist

with higher run idle voltage, the issue no longer exist

video bios has idle value at 800mv, sensors report it idle at 750mv (the issue occur)

video bios has idle value at 820mv, sensors report it idle at 775mv (the issue disappear)

compared RX480 and RX580 video bios,  (RX580 bios set idle value at 750mv) and (RX480 bios set idle value at 800mv)

suspect AMD video driver treats both RX480 and RX580 to idle at same 750mv , which likely is contributor of these random crash/freeze/blackout issues under idle/suspend/wake/low-load conditions

for AMD team,

please make video driver to respect RX480 video bios idle value of 800mv, do not run it at 750mv, 

above tests are evidence that it result in difference of  (stable system) vs (unstable system) 

[DasKänguru]

Whenever i edit the bios file accordingly, my AMD driver (newest release) will no longer work, did i do something wrong in polariseditor or do i need an older driver?