0

I'm writing a convnet using torch and cudnn and having some memory issues. I tried debugging the script with cuda-memcheck only to notice it actually runs when fed through cuda-memcheck (albeit slower than it should).

Turns out if cuda-memcheck is running in the background, a separate instantiation of the script itself also runs fine.

Any idea what might be happening here?

talonmies
  • 70,661
  • 34
  • 192
  • 269
Haggis
  • 17
  • 1
  • Have you tried running [this](https://github.com/soumith/cudnn.torch)? You should provide some more information about your "memory issues" and how you are using Torch. – Atcold Oct 20 '14 at 19:11
  • @Haggis this was a bug in NVIDIA Cudnn R2 release candidate 1. It has since been fixed in Cudnn R2 - rc2. – smhx Feb 03 '15 at 02:10

0 Answers0