> How is that possible? I'll guess: It's all in-memory, so not I/O-bound. Rows are processed in parallel with PGStrom (in 15MB chunks), while Postgres does it

> How is that possible? I'll guess: It's all in-memory, so not I/O-bound. Rows are processed in parallel with PGStrom (in 15MB chunks), while Postgres does it sequentially in one thread. Apparently moving the data around is fast enough not to matter.