Allocating large arrays on Tesla cards


I having difficulties in allocating large matrices on the Tesla (k40) cards beyond 15k x 15k dense matrices, while the memory of 12GB is more than enough to accommodate much larger matrices it throws a segmentation faults.

I would like to allocate and compute on matrices of at least 60k x 60k.

Is there a switch that I do not know about that need to include? or SM must be set to >= 3.5?

Has anyone has been successful in allocating such large matrices? If so, how?


Segmentation faults would tend to indicate a problem on the host side. So the problem may not be where you think it is.

For assistance with debugging, it is generally necessary to post concise, buildable, runnable, code that reproduces the issue.