Sirius: A GPU-native SQL engine

Summary

Sirius is a GPU-native SQL engine. It plugs into existing databases such as DuckDB via the standard Substrait query format, requiring no query rewrites or major system changes. Sirius currently supports DuckDB and Doris (coming soon), other systems marked with * are on our roadmap. Performance Running TPC-H on SF=100, Sirius achieves ~10x speedup over existing CPU query engines at the same hardware rental cost, making it well-suited for interactive analytics, financial workloads, and ETL jobs. Supported OS/GPU/CUDA/CMake Ubuntu >= 20.04 NVIDIA Volta™ or higher with compute capability 7.0+ CUDA >= 11.2 CMake >= 3.30.4 (follow this instruction to upgrade CMake) We recommend building Sirius with at least 16 vCPUs to ensure faster compilation. Dependencies (Option 1): Use AWS Image For users who have access to AWS and want to launch AWS EC2 instances to run Sirius, the following images are prepared with dependencies fully installed. AMI Name AWS Region AMI ID Sirius Dependencies AMI (Ubuntu 24.04) 20250611 us-east-1 ami-06020f2b2161f5d62 us-east-2 ami-016b589f441fecc5d us-west-2 ami-060043bae3f9b5eb4 Supported EC2 instances: G4dn, G5, G6, Gr6, G6e, P4, P5, P6. Dependencies (Option 2): Use Docker Image To use the docker image with dependencies fully installed: sudo docker run --gpus all -it yifeiyang7/sirius_dependencies:latest bash If encounting errors like the following when running the docker image as above: docker: Error response from daemon: could not select device driver “” with capabilities: [[gpu]]. This means nvidia-driver or nvidia-container-toolkit is not installed. To install nvidia-driver : sudo apt install nvidia-driver-535 To install nvidia-container-toolkit , please follow the instructions. Finally restart docker by sudo systemctl restart docker Dependencies (Option 3): Install Manually Install duckdb dependencies sudo apt-get update && sudo apt-get install -y git g++ cmake ninja-build libssl-dev Install CUDA If CUDA is not installed, download here. Follo...