Graph database intended for OLAP implementing Tinkerpop and designed to run on GPUs, specifically using mapd-core

I want to basically end up with a Titan/Janusgraph-like system backed by mapd-core. The graph framework should be able to leverage the features that mapd-core provides instead of simply using it to stash data. I do also want to expose features the CUDA ecosystem can provide via Gremlin to really leverage the GPUs involved. Also a few things regarding mapd-core, I want to make sure that users can enable either gpu+cpu or cpu-only mode as well as packaging up mapd-core so that users have an easier time building this project.

What I consider success for this project (or a production ready build) will be based off the performance stats of working with Graph500 data, specifically scales higher than 30 if possible. When it comes time to start testing, I'll include a performance test so others can see for themselves. Also, to ensure that results can be reproduced, I'll use EC2 instances to execute the tests for a clustered test and a GTX 1060 for a single node test.