Problems using GPU on Macbook Pro ... any ideas?


(WG) #1

I have a 2015 macbookpro (OS X Yosemite) with a NVIDIA GeForce GT 750M card.

Installed the latest version of CUDA (8.0) and followed the directions from here:

Getting this error when attempting to use the GPU:

WARNING (theano.sandbox.cuda): CUDA is installed, but device gpu is not available (error: Unable to get the number of gpus available: CUDA driver version is insufficient for CUDA runtime version)


(WG) #2

FIXED.

Problem: Installed the latest version of CUDA (8.0) when my driver required 7.5

Solution: Uninstalled 8.0 and then installed 7.5 (see #3 and #4 below), downloaded the proper version of cuDNN and copied the relevant files into the appropriate directories (under /usr/local/cuda), made sure my paths were set correctly in ~/.bash_profile … and voila.

Links to help others looking to develop locally on their OS X machines that aided me:

  1. http://blog.wenhaolee.com/run-keras-on-mac-os-with-gpu/
  2. http://philipperemy.github.io/configure_gpu_for_deep_learning/
  3. http://stackoverflow.com/questions/36942135/nvidia-cuda-toolkit-7-5-27-failing-to-install-on-os-x
  4. https://devtalk.nvidia.com/default/topic/520257/cuda-programming-and-performance/uninstall-cuda-in-mac-/post/4194565/#4194565

#3

@wgpubs how does the macbook gpu perform?


(WG) #4

@stickperson Not very well at all.

In fact, it was so poor that I pretty much switched to running everything on my Windows laptop.

Outside of some of the smaller notebooks in the course, I’d advise using the EC2 server or considering Windows.


#6

Here’s my setup http://adityagore.blogspot.com/2017/07/macbook-pro-setup-for-fastai-mooc.html


(taotaopan) #7

I have checked your post. Did you install it on virtual machine? since the bash file is for Linux I think.


#8

Sorry, just saw this. This writeup was for setting up on my MBP. No VMs just bare metal.