Shortest paths on huge graphs: Neo4J or OrientDB?

Kia Ora, I have a program that very frequently requires finding the fastest path (both the node sequence and total cost/length) on graphs containing ~50k nodes. Per run, I require on the order of millions of shortest path requests. I have just finished an OrientDB implementation which has significantly improved the compute time over my initial, non-graphDB attempt (which simply crashed). To perform testing, I am running the server locally on a series of distributed machines. However, in theory, would Neo4J, or another such platform, be faster again? What gains could I expect to receive? Could I host this process online, for example? Ngā mihi.