And even then, it really depends on what you're trying to do.

My use case a few years ago was such that _every_ dedicated graph engine I tried would just OOM on the test set of ~10M nodes, which I attributed to insufficient pruning. Postgres was able to handle it, and still worked on the production set of ~1.5B nodes. The SQL took some tuning, and wasn't _pretty_, but it did _work_ where nothing else did...