all CUDA-capable devices are busy or unavailable? Intermittently unable to access my GeForce 8600M G

Accelerated Computing CUDA CUDA Programming and Performance
1

I’ve got a MacBook Pro with an NVIDIA GeForce 8600M GT (128MB RAM).

I’m running CUDA 3.1, and when I try the bandwitdth test from the SDK samples, I get:

bandwidthTest.cu(600) : cudaSafeCall() Runtime API error : all CUDA-capable devices are busy or unavailable.

After I reboot, I am able to run the test, but I get an out of memory error at the end.

Several questions arise:

  1. What am I doing to get the card in a busy/unavailable state?

  2. Is there any way to reset the card without rebooting?

In addition, I tried to downgrade the developer driver from 3.1.17 to 2.3, just in case the driver was a problem, but when I do that, the NVIDIA control panel still reports I am running 3.1.17, even after a reboot.

Any guidance would be appreciated!

Thanks,
Dave

2

I’ve got a MacBook Pro with an NVIDIA GeForce 8600M GT (128MB RAM).

I’m running CUDA 3.1, and when I try the bandwitdth test from the SDK samples, I get:

bandwidthTest.cu(600) : cudaSafeCall() Runtime API error : all CUDA-capable devices are busy or unavailable.

After I reboot, I am able to run the test, but I get an out of memory error at the end.

Several questions arise:

  1. What am I doing to get the card in a busy/unavailable state?

  2. Is there any way to reset the card without rebooting?

In addition, I tried to downgrade the developer driver from 3.1.17 to 2.3, just in case the driver was a problem, but when I do that, the NVIDIA control panel still reports I am running 3.1.17, even after a reboot.

Any guidance would be appreciated!

Thanks,
Dave