pgrust 0.2 hits 300x Postgres on ClickBench with batching, operator fusion, and SIMD
pgrust 0.2 clocks roughly 300x Postgres on ClickBench — ahead of ClickHouse — and runs 30% faster than Postgres on OLTP, with the query engine rewrite accounting for about 10x of that gain. The speedup comes from three portable techniques: on a 500M-number sum, batching cut the Volcano model's 1.3s to 480ms, fusing adjacent operators to 358ms, and SIMD to 135ms. Benchmarks ran on an AWS c8g.4xlarge against PostgreSQL 18.4 with parallel workers disabled and data cached in shared buffers.
Source: malisper.me ↗
Many of the optimizations we've made target this. The query engine is the main user of CPU in a database.
pgrust developers
Why this matters
- → Demonstrates 300x speedup possible when query engines target modern CPU/memory bottlenecks instead of 1980s di
- → Batching, operator fusion, and SIMD are portable techniques applicable beyond pgrust
- → Shows why legacy database designs leave performance on the table despite decades of optimization effort
Query engines, reimagined