Fix CUDA driver version check.

[StefanOberhumer]

When accessing CUDAMiner, CUDAMiner::getNumDevices() is
called which can cause an exception.
Catch the exception, display the error message and exit.

Found this problem while searching solution for ISSUE #1378 - maybe this solves the problem.

[smurfy]

Well, while i like this change, why exit at all if cuda is not present?

[StefanOberhumer]

@smurfy

Well, while i like this change, why exit at all if cuda is not present?

If a user requests CUDA ( -U or -X ) I think we should exit if CUDA fails (currently ethminer "just" segfaults ;-) ).

[smurfy]

On -U, yes but on -X i do not know. For example i like to have just "one" commandline script for my amd and nvidia cards, so i would like to use -X, which exited in the past due to cuda error even if used on an amd only rig. (i think it still does)

I have nothing against the warning/error popup, just why exit?. While not everything is great what claymore does, the handling of mixed rigs/gpu detection and miner type is great with it.
But i think this is something for the future.

[StefanOberhumer]

On -U, yes but on -X i do not know. For example i like to have just "one" commandline script for my amd and nvidia cards, so ....

You're right - may be we can address this in the future.