8
$\begingroup$

NetTrain's option TargetDevice -> "GPU" works for old macbooks that have GEForce GT 750m cards (like mine), but what about other nvidia cards?

Does it (or will it) work with opencl based graphics card types?

$\endgroup$
  • $\begingroup$ I don't think we will ever support OpenCL in NetTrain, even Apple has deprecated it: "Apps built using OpenGL and OpenCL will continue to run in macOS 10.14, but these legacy technologies are deprecated in macOS 10.14." ~ developer.apple.com/macos/whats-new $\endgroup$ – Sebastian Jun 5 '18 at 11:28
  • 1
    $\begingroup$ @Sebastian Do you think you might use Metal or something similar as a replacement? $\endgroup$ – Carl Lange Jul 14 '18 at 8:03
  • $\begingroup$ @CarlLange: definitely not, unless Apple wrote a complete Metal backend for MXNet (which is very unlikely). Its a massive project to implement an efficient Metal backend for a feature of very limited utility. $\endgroup$ – Sebastian Jul 16 '18 at 21:53
  • $\begingroup$ @Sebastian Fair enough. Its a pity that many of us will have to be left out of the machine learning race because of Apple deprecating OpenCL. Ah well. Thanks for your answer! $\endgroup$ – Carl Lange Jul 16 '18 at 23:22
  • $\begingroup$ @CarlLange: you still have the option of using cloud NVIDIA GPUs (eg on EC2), which have GPUs appropriate for training modern nets, like V100s (laptop GPUs which the OP was asking about are pretty hopeless for anything more than toy examples). Making it easier to train on EC2 is something we are looking into. $\endgroup$ – Sebastian Jul 17 '18 at 0:37
10
$\begingroup$

All NVIDIA GPUs except newest 10xx.

http://community.wolfram.com/groups/-/m/t/917616

"A number of TensorFlow users are reporting problems using a 1080/1070 with CUDA v7.5 and cuDNN v5.0 (which we are currently using for 11.0)."

"So it looks like you will need to wait for 11.1, which will definitely support CUDA 8.0."

OpenCL is in plans.

https://github.com/dmlc/mxnet/issues/621

Mathematica's neural network functionality is based on MXNET. Wolfram developers are contributing to this project very actively.

https://github.com/dmlc/mxnet/search?q=sbodenstein&type=Issues https://github.com/dmlc/mxnet/search?q=taliesinb&type=Issues

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.