Agreed. They brush aside the years of high performance computing graph implementations, eg, cuStinger. If you look at systems like neo4j's GDS and how other mul

Agreed. They brush aside the years of high performance computing graph implementations, eg, cuStinger. If you look at systems like neo4j's GDS and how other multipurpose systems are going wrt views/projections for accelerated compute, that has been enabling targeting way more performance without dying under complexity. Benchmarking perf without that kind of comparison is weird, I'm surprised reviewers allowed that. (The work may still be good.. just you can't know without that.)