Unspecified Launch Error - No Cuda Device


We have developed computer vision application that is using CUDA and it runs fine for ~6 hours or so until
we get an “Unspecified Launch Error”. This seems to be completely random and it occurs
on Ubuntu 9.10 and/or Windows XP using 2.3 toolkit and the latest drivers.

I’m using the following cards:

GeForce 8600 GT (Windows and Linux)
GeForce GTX 260 (Windows only)
ION - Linux

I’ve tried to kill my running thread using cudaThreadExit() and reinitialize/purge all memory but this seems to only work
a few times before we start getting the second error of “No CUDA Capable Device is Present”. It is critical
that this application be up 24x7x365 as it is meant for surveillance application.

I have been ALL over the code looking for possible segmentation faults or writing past array boundaries but this doesn’t seem
to be the problem.

Any thoughts would be appreciated!


No one has any thoughts?