https://www.sirius-db.com/

Sirius Logo

Sirius

SQL Analytics at GPU Speed

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. By offloading query execution
to GPUs, Sirius achieves over 10x speedup at the same hardware rental
cost!

Email Us
We'd love to hear from you -- questions, contributions, or
collaborations.
Join Slack
Connect with the community, ask questions, and stay updated.
GitHub
Source code available on GitHub

Key Features

GPU-Native Speed

Sirius is 10x faster than DuckDB and 60x faster than ClickHouse at
the same hardware rental cost. Built for GPUs, Sirius targets 100x
speedups over CPU-based SQL engines.

Seamless Integration

Sirius plugs into databases like DuckDB and Doris via Substrait,
accelerating SQL workflows without changing your stack. It supports
CPU fallback for full compatibility.

Deploy Anywhere

Sirius supports a wide range of deployment options, from cloud to
on-prem, delivering GPU-accelerated performance wherever you run.

Architecture

Sirius Architecture Diagram

Systems marked with * are on our roadmap

Publications

  * Rethinking Analytical Processing in the GPU Era
    Bobbi Yogatama*, Yifei Yang*, Kevin Kristensen, Devesh Sarda,
    Abigale Kim, Adrian Cockcroft, Yu Teng, Joshua Patterson, Gregory
    Kimball, Wes McKinney, Weiwei Gong, Xiangyao Yu
    arXiv, 2025
    * Equal Contribution
  * Debunking the Myth of Join Ordering: Toward Robust SQL Analytics
    Junyi Zhao, Kai Su, Yifei Yang, Xiangyao Yu, Paraschos Koutris,
    Huanchen Zhang
    SIGMOD, 2025
  * Accelerate Distributed Joins with Predicate Transfer
    Yifei Yang, Xiangyao Yu
    SIGMOD, 2025
  * GPU Databases--The New Modality of Database Analytics
    (presentation)
    Bobbi Yogatama, Xiangyao Yu
    HPTS, 2024
  * Scaling your Hybrid CPU-GPU DBMS to Multiple GPUs
    Bobbi Yogatama, Weiwei Gong, Xiangyao Yu
    VLDB, 2024
  * Predicate Transfer: Efficient Pre-Filtering on Multi-Join Queries
    Yifei Yang, Hangdong Zhao, Xiangyao Yu, Paraschos Koutris
    CIDR, 2024
  * Accelerating User-Defined Aggregate Function (UDAF) with
    Block-wide Execution and JIT Compilation on GPUs
    Bobbi Yogatama , Brandon Miller, Yunsong Wang, Graham Markall,
    Jake Hemstad, Gregory Kimball, Xiangyao Yu
    DaMoN@SIGMOD, 2023
  * Orchestrating Data Placement and Query Execution in Heterogeneous
    CPU-GPU DBMS
    Bobbi Yogatama, Weiwei Gong, Xiangyao Yu
    VLDB, 2022
  * Tile-based Lightweight Integer Compression in GPU
    Anil Shanbhag*, Bobbi Yogatama*, Xiangyao Yu, Samuel Madden
    SIGMOD, 2022
  * A Study of the Fundamental Performance Characteristics of GPUs
    and CPUs for Database Analytics
    Anil Shanbhag, Samuel Madden, Xiangyao Yu
    SIGMOD, 2020