The Journey to Zero-Copy: How chDB Became the Fastest SQL Engine on Pandas DataFrame

Auxten Wang Technical Director at ClickHouse

In this talk, Auxten Wang, creator and lead developer of chDB, shared the story of how the world’s fastest OLAP engine was made to fly on Pandas DataFrames, reducing query times from 30 seconds to just 0.5 seconds. He walked through chDB’s evolution from a serialization-bound prototype into a fully zero-copy SQL engine for Pandas, covering how DataFrame and Parquet round-trips were eliminated, how Python GIL limitations were overcome, and how string handling was re-thought for parallel performance. The result is true DataFrame in and DataFrame out SQL with ClickHouse speed, delivering up to 200× faster analytics, native JSON support, streaming for data larger than RAM, and a complete zero copy loop with no servers, no setup, and no friction.