Last week we launched pgrust 0.2, a release that detoxified performance. It’s 10x faster than the prior version, 30% quicker than Postgres on OLTP workloads, and astonishingly 300x faster than Postgres on Clickbench, even outpacing Clickhouse.

The gains stem from destaque changes. Postgres, built in the 80s, was designed for disk‑bound workloads. Today, most data fits in RAM and CPU/memory bandwidth drive speed, so a modern engine can shave overhead that Postgres still carries.

Take a simple SUM query: Postgres takes ~20 seconds on a ومس 20‑core machine, while a raw Rust loop finishes in 358 ms—roughly 55x faster. The difference comes from eliminating locking, parsing, and other Postgres internals.

The core speedup is in the query engine. Postgres uses a Volcano model that processes one row at a time, incurring heavy function‑call overhead. By batching 1024 rows at once and fusing operators into single nodes, the batched engine drops runtime from 1.3 s to ~480 ms, approaching the raw loop.

These optimizations—batching, operator fusion, and SIMD—show that a lightweight, Rust‑based engine can leap ahead of legacy systems.