Strange freezes with Tesla C2050 - Help needed! Help needed!!!!

Thanks for keeping me updated!

In my case it’s also a university computer, that’s why I’m sticking to a computer repair service, they wouldn’t like it if I would be messing with the hardware.

If the card is going to work alone in the computer, than we are going to replace the GTX295 with a low-end Geforce. If not, than I will also test it in my home computer (a much worse one, but with the Tesla alone the power should be enough). If this is neither working than we will test it in our next computer (coming in 1-2 months). That one will have a 1200 W supply which should be more than enough (besides the two serves-processor).

Thanks for keeping me updated!

In my case it’s also a university computer, that’s why I’m sticking to a computer repair service, they wouldn’t like it if I would be messing with the hardware.

If the card is going to work alone in the computer, than we are going to replace the GTX295 with a low-end Geforce. If not, than I will also test it in my home computer (a much worse one, but with the Tesla alone the power should be enough). If this is neither working than we will test it in our next computer (coming in 1-2 months). That one will have a 1200 W supply which should be more than enough (besides the two serves-processor).

Have you been able to get these issues resolved? Anyone else seeing problems with Tesla 20-series cards failing?

Have you been able to get these issues resolved? Anyone else seeing problems with Tesla 20-series cards failing?

Not yet…

Not yet…

How are you switching between GPUs to run the SDK samples? By default, without re-building, they run on device 0.

How are you switching between GPUs to run the SDK samples? By default, without re-building, they run on device 0.

I do not remember exactly, but either in the DirectCompute or the OpenCL version the GPU can be changed (but I won’t try it now through network connection, because than I would need to go in and stop the frozen computer). Also, in Sisoft Sandra tests the GPU ID can be changed.

Otherwise CUDA nbody was running automatically on Tesla, after I have added the new card. Also my program has listed it as the GPU ID 0. I have also tried my program (CFD) on it, the result was an instant freeze with blue death (just for comparion on my home GTX460 it was working perfectly and really fast). Also we have been using it on the GTX295 for months without problems.

I do not remember exactly, but either in the DirectCompute or the OpenCL version the GPU can be changed (but I won’t try it now through network connection, because than I would need to go in and stop the frozen computer). Also, in Sisoft Sandra tests the GPU ID can be changed.

Otherwise CUDA nbody was running automatically on Tesla, after I have added the new card. Also my program has listed it as the GPU ID 0. I have also tried my program (CFD) on it, the result was an instant freeze with blue death (just for comparion on my home GTX460 it was working perfectly and really fast). Also we have been using it on the GTX295 for months without problems.

Most, if not all, of the examples support an optional -device= flag that you can use to specify the device at runtime.

Most, if not all, of the examples support an optional -device= flag that you can use to specify the device at runtime.

Nothing yet from my local supplier, still waiting…

Nothing yet from my local supplier, still waiting…

By any chances, Are they pre-production cards?

By any chances, Are they pre-production cards?

I’m not sure what you mean with “pre-production cards”; mine is obtained through “Mad Science” program and it has “EDU” suffix in its part number.

I’m not sure what you mean with “pre-production cards”; mine is obtained through “Mad Science” program and it has “EDU” suffix in its part number.

If you don’t know, you don’t have one. :) NVIDIA sometimes seeds prototype cards (usually made in the last round before official production) to external developers for testing.

If you don’t know, you don’t have one. :) NVIDIA sometimes seeds prototype cards (usually made in the last round before official production) to external developers for testing.