Aggregate effective memory bandwidth if you use the CPU cache well. Graph databases are not compute intensive, nor are they particularly large data models, but they are extremely memory I/O intensive. The classic model for graph-oriented HPC was vast numbers of weak cores and barrel processors for this reason, though the utility of specialized hardware architectures has been greatly reduced by better software architecture for this purpose over time.
Fraud analytics.
You're looking for patterns across large numbers of entities and relationships.
And ideally you want this all done in real-time so you can stop transactions before they are approved.
Any chance you could expand on this? I hear about pattern matching for fraud analytics but I have never seen concrete examples, or even anything in the literature.
A social media service for hundreds of millions of users?