1

Caffe is a deep learning framework that also includes a Matlab interface (Matcaffe). While Caffe supports multiple GPUs it seems that Matcaffe does not at the moment. Does anyone know of a workaround?

Daniel
  • 36,610
  • 3
  • 36
  • 69
ASML
  • 199
  • 12
  • Could you give an example which is using multiple gpus in caffe while not in matcaffe? To my knowledge matcaffe can use multiple gpus for different tasks, which is all caffe can do with multiple gpus. – Daniel Oct 31 '15 at 00:52
  • 2
    In Caffe, we can do './caffe train [...] -gpu all' to train a CNN on all available GPUs. In Matcaffe, there's only 'caffe.set_device(gpu_id);'. While this let's me choose which GPU to use, I cannot use them in parallel. Specifically, I have 20k images and would like to distribute the forward pass through the net to 64 GPUs. – ASML Oct 31 '15 at 01:14
  • Thanks, that explains your question. Did not know caffe supports using multiple gpus in the same training process now. – Daniel Oct 31 '15 at 10:31
  • There is no issue regarding this limitation opened at github. Maybe you could create one https://github.com/BVLC/caffe/issues – Daniel Oct 31 '15 at 10:40
  • related: http://stackoverflow.com/questions/33726333/caffe-pycaffe-set-all-gpus – Shai Nov 16 '15 at 08:21

0 Answers0