I am running a proof of concept CUDA application on a multi-GPU environment (nvidia-smi -L):
GPU 0: Tesla C2050 / C2070 (S/N: 0321310045770)
GPU 1: Tesla C2050 / C2070 (S/N: 0321510020975)
GPU 2: GeForce GTX 480 (UUID: N/A)
GPU 3: GeForce GTX 295 (UUID: N/A)
GPU 4: GeForce GTX 295 (UUID: N/A)
Nevertheless I am running in a single card. The problem is that I get different execution times when running
on card ‘0’/card ‘1’ (50% slower for card 1). Both cards are the same model and all information obtained through
nvidia-smi -q -i gives exactly the same information for both. What could be going on?.
Thanks in advanve.