You can start with the free trial, which comes with $300 of free credits. Note 4: The installation is quite involved, but if you save a disk image once things are working you'll only have to do it onceĪlright, onwards: Set up a virtual machine on Google Cloud Platform Get an accountįirst make a Google Cloud account here if you don't already have one. Note 1: Remember to stop your instance you're done with it so you don't continue getting charged Note 2: Setting up your GPU quota can take a day or two, so plan accordingly Note 3: This tutorial uses Python 3, TensorFlow 1.6.0, CUDA 9.0, and CuDNN 7.0.4 for CUDA 9.0Īnd all software versions must be compatible with each other for the install to be successful, so only change them if you know exactly what you're doing.
It will guide you through the minimal steps to spin up a GPU-enabled VM in the Google cloud, install the latest version of TensorFlow (1.6.0 as of ), and get it to recognize the GPU and rapidly train a simple network.
This is an updated and streamlined version of this excellent tutorial by Steve Domin. INSTALLING AND TESTING GPU-ENABLED TENSORFLOW ON A GOOGLE CLOUD PLATFORM VIRTUAL MACHINE
