graphlab.neuralnet_classifier.create() not using GPU

User 5340 | 7/9/2016, 11:57:15 PM


I recently got a Ubuntu 16.04 machine with NVIDIA GeForce GTX 970M GPU. I'm using CUDA 7.5. I installed graphlab with GPU acceleration as outlined here, and ran the MNIST tutorial code from the user guide to see if the GPU acceleration is working. Unfortunately, graphlab.neuralnet_classifier.create() still says "Creating neuralnet using cpu" when I run it. How can I get it to use the GPU instead?

Thanks, Beata


User 5340 | 7/14/2016, 3:47:10 AM

It seems like the GPU is being used properly when I'm training a convolutional neural network, but not with the built-in "regular" MNIST network. Is that by design?

User 940 | 7/18/2016, 6:53:23 PM

Hi @Beata,

Unfortunately, I'm unable to reproduce this behavior. Could you share the code you are running, along with its output, for both the case where the GPU is and is not used? This will help us debug.

Cheers! -Piotr