11-08, 15:20–16:00 (US/Eastern), Central Park West
Polars is a high performance query engine with a DataFrame frontend. This year the project has landed its 1.0 milestone and this September it has landed GPU acceleration in collaboration by NVIDIA Rapids. The combination of the Polars query optimizer and the NVIDIA acceleration, makes it very suitable for even higher performance workloads.
In this talk we explain the integration of GPU acceleration in Python Polars with cuDF. We'll explore the tightly-coupled approach where cuDF hooks in after the Polars IR (Intermediate Representation), allowing for seamless GPU execution on top of Polars' existing optimizations. Learn about the design constraints, implementation challenges and performance on different query types. Finally, we’ll explore what’s more to expect from the Polars team.
Polars is a high performance query engine with a DataFrame frontend. This year the project has landed its 1.0 milestone and this September it has landed GPU acceleration in collaboration by NVIDIA Rapids. The combination of the Polars query optimizer and the NVIDIA acceleration, makes it very suitable for even higher performance workloads.
In this talk we explain the integration of GPU acceleration in Python Polars with cuDF. We'll explore the tightly-coupled approach where cuDF hooks in after the Polars IR (Intermediate Representation), allowing for seamless GPU execution on top of Polars' existing optimizations. Learn about the design constraints, implementation challenges and performance on different query types. Finally, we’ll explore what’s more to expect from the Polars team.
No previous knowledge expected
Ritchie Vink is the Author of the Polars DataFrame library and Founder/CEO of the Polars company. Before Polars,, he was working for 5 years in machine learning and software engineering, after he made the switch from Civil Engineering.