Groute is a multi-GPU programming model and framework, implemented over CUDA. The framework is designed to optimize the implementation of irregular algorithms over a single node with multiple GPUs, and does so with an asynchronous approach.
Groute is the framework behind our PPoPP 2017 paper [paper][pres], and is still under active development.
You can submit contributions via a GitHub pull-request. Any other comments/requests/bug-reports are welcome, please contact us at msutton@cs.huji.ac.il or talbn@cs.huji.ac.il