multiple threads in one GPU

[jean]

Hi,

i ran multiGPU.c example and it all went good. I have 4 GPU so it created 4 threads, one for each GPU. Then i changed to
numThreads = numDevices * 2;
so it would create 2 threads per GPU. But when i run it, the 2 threads in GPU 0 run smoothly, but in the other 3 GPUs, 1 of the threads give me error:
Thread 5 - CUBLAS error: not initialized

Also, what i want to do is to define how many cores are used in a cula function in one GPU, by defining the number of threads in a cula function, supposing that each thread goes for one core. In the multiGPU.c example, when i put 2 threads per GPU, it does 2 independent cula function in each GPU.

Can i get help?
Thank you

[john]

Can you tell me which version of CULA you are using? So far I have only managed to reproduce this on an older version of CULA.