GeForce 295GTX execution divergence

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello

I’m executing the same application on a geForce 9800GT (with one device) and on a geForce 295GTX (two devices).
The results of the application are good if it’s done on the 9800GT but not good from de 295GTX. In fact, I get random ‘unspecified launch failures’ errors during the execution in the 295GTX.

Does anybody knows why this can be happening?

Any issue about I have to take care if I the execution is done on a two devices gpu gard? Any compilation issue?

Do you know if the compilation spreads the global variables between the two devices (memory allocations) ? And if it’s affirmative, Does the global device memory of one device is visible by the GPU of the other device?

Thank you!