xinerama, 1 cuda card and interop problems

Using Cuda 3.0 beta.

If i have two graphics cards:
#1 8800 GTX: Xinerama, 2 monitors
#2 8800 GT: Dedicated cuda

Can I use Opengl interop between them? When I try after a period of time there is a CUDA_LAUNCH_FAILURE and a kernel message:
[23696.782542] NVRM: Xid (0001:00): 13, 0003 00000000 000050c0 00000368

The same program using only card #1 for everything works fine.

Code 13 errors are pretty grave, usually indicated that either the driver or card has just suffered brain death. If you have a full repro case, you might want to post it, because you have either just discovered a driver bug, or one of your cards is flakey. I had a DDR3 9500GT in a development box that used to do that when running very memory intensive kernels, eventually I put it down to bad memory. On every other CUDA capable device I could get my hands on, the same code worked fine.

Thanks for your experience. It made me determine that one of my graphics cards is dying. I tested:

8800 GTX: cuda and display : works

8800 GT: cuda and display: fails

8800 GTX display, GT cuda: fails

8800 GT display, GTX cuda: works.