Pages: [1]
Oink-NL
 
BAM!ID: 65419
Joined: 2009-02-11
Posts: 121
Credits: 391,397,626
World-rank: 4,175

2009-10-17 14:12:54

Most frustrating thing on the world: BOINC is running happily all day: but as soon as you leave, a VPU recovery occurs and all ATI tasks keep hanging untill someone (or something) restarts BOINC. I found out that it is pretty stable when no user is logged on, but incidentally still VPU recovers.

My system:

  1. Core2 Duo 8400
  2. ATI Radeon 4850
  3. 4GB memory
  4. 1TB HDD


Software:


  • Windows XP Pro x64
  • Boinc 6.10.13 n(Running as a service)
  • Ati 9.7 drivers (including CCC)



I use the windows command eventtriggers.exe to monitor the ACE Even Log (Which is created by the CCC installation). Unfortunately it only shows information messages when it crashes, no errors. I ran the following command from the command prompt as administrator:

Code:
eventtriggers /create /TR "Restart Boinc" /TK C:\rb.cmd /so ACEEventLogsource /ru system


This line creates a background process monitoring the event log. In case of activity (in this case ANY activity) from source ACEEventLog, it calls C:\rb.cmd, which contains the following code:

Code:
net stop BOINC
ping 127.0.0.1
net start BOINC
exit


In short: stop Boinc service, wait 4 seconds (That's what the ping does. I know it's not the way, but it's quick and dirty, but above all: effective), then starts the boinc client again.

This enables me to tweak settings (overclock GPU, change settings in MW xml file) to speed things up a bit.

HOWEVER: keep an eye on the log file. ATI/CCC seems to disable the driver if too many VPU recovers occur. The only way out of that is a restart.

I hope somebody can use this tweak to get his GPU to do some work without having to monitor the system constantly.

Greetz,

Oink
Guest

2009-10-17 15:09:44

hi frank,

Oink-NL wrote:

  • Windows XP Pro x64
  • Boinc 6.10.13 n(Running as a service)
  • Ati 9.7 drivers (including CCC)


  • 1st idea: at least with NVIDIA GPU's running boinc as service is a bad idea.

    2nd one (i got no ATI's around for years): why not get rid of the CCC and use rivatuner or something else?


    gruss, frank.
    Oink-NL
     
    BAM!ID: 65419
    Joined: 2009-02-11
    Posts: 121
    Credits: 391,397,626
    World-rank: 4,175

    2009-10-17 15:55:22

    frankhagen wrote:

    1st idea: at least with NVIDIA GPU's running boinc as service is a bad idea.

    2nd one (i got no ATI's around for years): why not get rid of the CCC and use rivatuner or something else?


    gruss, frank.


    Guten Tag Frank,

    1. nvidia has nu VPU recover, because it's more stable. CUDA apps are developed generally with some help from NVidia. Stream Applicatio developers have to figure it all out by themselves. Hence the idea of figuring something out to counteract the instability, at least until the apps are more stable (It's All Beta at best). System managers don't want users to control BOINC, so having it run as a service is an advantage for people having access to a large number of machines. (By the way, at work my own workstation is running BOINC as a service with NVIDIA. no problems for me there)

    2. CCC enables VPU recovery and writes events in the eventlog. That combination enables the system to restart BOINC. The CCC systray applet doesn't need to be running and is in my experience one big cause for VPU recover events. It's all con's and pro's. Off course everybody can do what they want with this info .

    Greetz,

    Frank
    Pages: [1]

    Index :: Optimized clients :: Restart BOINC after VPU recover (ATI)
    Reason: