View Single Post
Old 03-27-11, 09:06 AM   #12
Chris M
Registered User
 
Join Date: Dec 2010
Posts: 16
Default Re: Quadro FX 2800M - lockups, resets, ...

Hallo danix
Quote:
Originally Posted by danix View Post
Have you found a way (preferably automated) to trigger the crash on demand?
Sadly not.
Quote:
Originally Posted by danix View Post
Based on the suspicion that PowerMizer is involved, we tried running the following on our M6500 overnight to see if it would crash, but it didn't:

Code:
while true; do nvidia-settings -a [gpu:0]/GPUPowerMizerMode=$[$RANDOM%2]; sleep 2; done
I tried that line for a short period of time. But as the PowerMizer mode switch itself isn't causing the troubles I've enhanced that to:
Code:
start=$SECONDS;count=0;count0=0;count1=0;while true; do count=$((count+1));mode=$((RANDOM%7<1));count0=$((count0+(mode==0)));count1=$((count1+(mode==1)));echo $count: $((SECONDS-start)) Seconds after start. Mode 0: $count0, Mode 1: $count1. Temperature `nvidia-settings -t -q [thermalsensor:0]/ThermalSensorReading` C;nvidia-settings -a [gpu:0]/GPUPowerMizerMode=$mode; sleep 10; done
This will stay mostly in adaptive mode and sometimes kick in to performance mode. So the frequencies will change as much as possible.
That run fine so I stopped it, the last lines were:
Quote:
1726: 17609 Seconds after start. Mode 0: 1486, Mode 1: 240. Temperature 48 C

Attribute 'GPUPowerMizerMode' (obiwan:0[gpu:0]) assigned value 0.
I.e. after nearly 5 hours running idle and only switching frequencies I couldn't force a hang up. (Temperatures stayed between 48C and 53C)

Thinking of it, it seems to me that mostly (allways?) the hang ups and crashes happen during user interaction. A very typical setup is that I'm reading a web page (i.e. everything is idle) and are switching to a different program (e.g. email client) that might even be on a different screen. I.e. the xserver has to handle mouse events and a graphic intense workload out of an idle state.
During long periods of leaving the laptop in an idle state hardly any (or even "no"?) hang ups occured.
Chris M is offline   Reply With Quote