BlazingSQL is a GPU-accelerated SQL engine that runs on It’s a comprehensive collection of configurable primitives. These primitives are meant to be operated on distributed data frames. BlazingSQL includes an I/O layer for interacting with different file systems, including S3 or HDFS. The SQL engine contains primitives that use different backends, such as UCX and TCP.

Project Background

  •  Platform: BlazingSQL 
  • Author: Rodrigo Aramburu
  • Released: August 2021
  • Type: Distributed SQL engine
  • License: Apache License 2.0
  • Supports: GPU, CUDA, and Python
  • Language: C++, Java, Python 
  • GitHub: BlazingDB/blazingsql


  • Industry solutions
  • Urban planning
  • Data science solutions


  • BlazingSQL uses the interface through its python library blazingsql. 
  • When someone runs an SQL query on this platform, it gets converted into relational algebra. That’s all because it leverages Apache Calcite.
  • In the BlazingSQL Core engine, Apache Calcite produces relational algebra. This physical, relational algebra plan becomes a directed acyclic graph (DAG). In DAG, every node is a kernel, and the connecting edges are called caches. 
  • Other support components of BlazingSQL include memory management, communication library, Interops, data structures, and I/O modules
Scroll to Top