JCuda

Databricks GPU clusters clusters support low-level CUDA development using Java libraries such as JCuda.

Launch a GPU enabled cluster with the preloaded Nvidia Libraries. Then download the JCuda libraries from http://jcuda.org/

You will need the JCuda Java Jar and the native library. Attach these libraries to your cluster.

This tutorial inspired by the classic Pi Estimation example at http://spark.apache.org/examples.html