GPUs are a great resource of computing power but yet not accessible from Apache Spark. We present a RDD implementation we've open sourced to leverage GPU computing power with Spark. We'll share the experiences we gained along the way implementing the RDD, and a real-world application using the RDD: What's the best way to bridge from Java to GPU code (OpenCL or CUDA)? From an architectural perspective - what's the best way to integrate a GPU processing facility into Spark? How much faster are typical Spark actions when using the GPU? What Spark actions are best processed on a GPU? Java-to-GPU bridges, best way to integrate GPU processing into Spark and performance evaluation.