My graphics driver crashes at random, it happened twice now, it recovers right away but causes the CUDA wu that is working at the time to error out. I have been told that my driver is old, I cant update it at this time however. Any ideas on how to prevent this?

I forgot to state that the driver crashes when the screen saver comes on.

Don't use the screen saver with CUDA. When your GPU is processing tasks, it cannot be used to display the graphics too.

---

I've got another crashing system:

STOP 0x000000EA THREAD_STUCK_IN_DEVICE_DRIVER

is the error message that comes out of the crash when I report the incident to Microsoft.

They say the "fix" is to turn off hardware acceleration.

I've got the lastest NVidia driver running my GeForce 8600 GT.

Any ideas out there? What's going to happen if I shut down hardware acceleration? Does everyting get real slow to the point I shouldn't be running BOINC?

Also, changing versions hasn't changed the symptoms. It's been doing it for a while.

Bruce Parker
Dumfries, VA

---

This nasty bug has about a hundred different causes. What's actually happening is that your video driver has essentially entered an infinite loop because your video adapter has locked up.

You could try:
- A better power supply, in case your videocard doesn't get enough power from your current one.
- Older drivers if this problem occurs with the latest drivers.
- Check your card's seating. Is it properly pressed into the AGP, PCI or PCIe slot? Else remove it and re-seat it.
- Make sure both your card and your computer have good cooling.
- Check for newer chipset drivers for your motherboard. Your motherboard manufacturer web site can help you along there.
- If your motherboard has an on-board Ethernet adapter, try disabling the "PXE Resume/Remote Wake Up" option in your system BIOS.

As for hardware acceleration, that's primarily used for Direct 3D/OpenGL gaming and watching video. You then use the GPU to calculate where pixels need to be, what their color is etc. instead of leaving that to the software to do. it's not used for CUDA.

---