(almost) without fail, if my computer is left for ten minutes and the display goes to sleep (I set my hard drives to never spin down though), and is then woken up by a mouse move of key stroke, an Nvidia driver error appears from a task bar icon.

I'd like to keep running cuda projects but I don't know that this isn't causing some kind of damage so am tempted to prevent GPU usage.

Is this problem known and is there a solution?

Thanks

my system if it's useful...
xeon quad core
16gb ddr3
nvidia 9800gt 1gb
60gb ssd os + applications
6tb raid 5 data (inc boinc data)
win 7 64 pro

What Boinc version? Which Cuda Projects? Which Nvidia drivers? Best to ask at the Project concerned, it's likely that it's their Cuda app or screensaver that will have caused the driver to restart.

Claggy

Latest Boinc version (6.12.33)
Latest Nvidia driver (285.62)

I don't know how to find out which project causes the crashes, but given the frequency of it it seems like it's any that are GPU accelerated.

You can only go test these things by changing your videocard driver to a lower one, and a next lower one, until you hit upon one that's stable for your system.

When you never find a stable driver, it may well be that your videocard is the culprit, that when it gets hot that it can't take it anymore and that it crashes. Nothing to do against that, but for either not use it for this, or replace it with a new (and better) one.

Make sure that your Windows screen saver is disabled (set to None). On non-Fermi cards the GPU may not like it that this thing comes on, even if you have told Windows to turn the monitor off.

And one last thing on this, if all is well, Windows will record why the videocard driver crashed. It will do so in Start->Administrative tools->Event Viewer. Somewhere in there there will be a log as to what crashes the driver.

Apropos, the latest recommended BOINC is 6.12.34.
The latest 6.12 BOINC is 6.12.41.
The latest test BOINC is 7.0.7.

Just to give an idea that 'the latest' isn't always the latest and that we prefer specific numbers, instead of 'the latest' which could be anything.

---

Sorry Boinc version is 6.12.34

Thanks for the event viewer tip, it reports post crash that "Display driver nvlddmkm stopped responding and has successfully recovered."

It also says...

#############################################

- System

- Provider

[ Name] Display

- EventID 4101

[ Qualifiers] 0

Level 3

Task 0

Keywords 0x80000000000000

- TimeCreated

[ SystemTime] 2012-01-12T18:13:25.000000000Z

EventRecordID 109406

Channel System

Computer Squid

Security


- EventData

nvlddmkm

###########################################


Is that of any use to anyone? If no one else is having this problem and can defuse it I might just disable GPU usage completely.

Thanks!

Latest Boinc version (6.12.33)
Latest Nvidia driver (285.62)

I don't know how to find out which project causes the crashes, but given the frequency of it it seems like it's any that are GPU accelerated.

So which GPU projects are you running?

270.xx and later drivers introduced 'asynchronous execution', unfortuently the drivers freakout when they are terminated abruptly, eithier causing the GPU to downclock, or the driver to restart,

The Solution is to use 267.xx or ealier drivers, or only run GPU projects that have introduced threadsafe Cuda apps,
the third option is to not interupt the GPU by using the 'Suspend GPU work while computer is in use?' option,

For more information on this problem have a read of this thread:

Recent Driver Cuda-safe Project List

Claggy

excellent! thank you! :D