Vote count:
0
I'm remotely running code on a GTX Titan Black and getting memory errors or outright crash+reboots despite there always being ample memory left.
I get 3 kinds of errors on different alterations of my code (the alteration being whether to load small or large batches of images on the GPU at each iteration):
- cuda runtime error : an illegal memory access was encountered
- cuda runtime error : out of memory
- shell freezes, impossible to ssh onto it for a few mins, so I assume GPU crashes and system reboots.
Is this a generic problem that other have run into? I'm aware it might be specific to my code since error varies on how I implement it, but maybe these unjustified memory errors also point to sys admin or driver restrictions?
Ubuntu 14.04, CUDA v6.5 (but got same errors with Ubuntu 13.04 and a previous version of CUDA), GTX Titan Black 6GB
GTX Titan Black memory errors
Aucun commentaire:
Enregistrer un commentaire