Use GPUs with R
These are the sample code and installer mentioned in my GPU-Powered Computing for Data Science with R Notebooks on Google Cloud’s AI Platform Notebooks [DC91511] talk [recording] at Nvidia GTC 2019. The talk covered various ways to speed up your data analysis for AI and ML workflows, with a focus on optimizing GPU usage
This repository contains consists of installation scripts to easily install R on Jupyter Notebooks (like AI Platform Notebooks).
The notebooks contain code that shows how easy it is to perform various Machine Learning/Deep Learning actions on CPU based notebooks versus GPU based notebooks, and you can use the installation script to add R support to any of the existing AI Platform Notebooks
This blog post describes the installation script in more detail: https://zainrizvi.io/blog/using-gpus-with-r-in-jupyter-lab/
Installation
To use the provided scripts on your AI Platform Notebooks, create a notebook VM and then run one of the below commands based on whether or not your notebook VM has GPUs attached. (Don't like running unknown code from the internet? I explain what they are doing in this blog post)
With CPUs only: 'sudo -- sh -c 'wget -O - https://raw.githubusercontent.com/ZainRizvi/UseRWithGpus/master/install-r-cpu.sh | bash'
With GPUs: 'sudo -- sh -c 'wget -O - https://raw.githubusercontent.com/ZainRizvi/UseRWithGpus/master/install-r-gpu.sh | bash'
Now, those scripts take a while to run. Instead, you can just use the containerized versions of AI Platform Notebooks, which come with Tensorflow 2 support built in.
Here are their repositories on docker hub:
- zainrizvi/deeplearning-container-tf2-with-r:latest-cpu
- zainrizvi/deeplearning-container-tf2-with-r:latest-gpu
And you can use the above custom containers to have a notebook running on AI Platform Notebook in just a couple minutes!