Rethinking Analytical Processing in the GPU Era DOI: XXXXXXX.XXXXXXX Conference: The Conference on Innovative Data Systems Research; January 18–21, 2026; Chaminade, USA ISBN: 978-1-4503-XXXX-X/2018/06
Bobbi Yogatama 1 ∗ , Yifei Yang 1 ∗ , Kevin Kristensen 1 , Devesh Sarda 1 , Abigale Kim 1 , Adrian Cockcroft 5 , Yu Teng, Joshua Patterson 2 , Gregory Kimball 2 , Wes McKinney 4 , Weiwei Gong 3 , Xiangyao Yu 1 · 2025
Read Rethinking Analytical Processing in the GPU Era DOI: XXXXXXX.XXXXXXX Conference: The Conference on Innovative Data Systems Research; January 18–21, 2026; Chaminade, USA ISBN: 978-1-4503-XXXX-X/2018/06 by Bobbi Yogatama 1 ∗ , Yifei Yang 1 ∗ , Kevin Kristensen 1 , Devesh Sarda 1 , Abigale Kim 1 , Adrian Cockcroft 5 , Yu Teng, Joshua Patterson 2 , Gregory Kimball 2 , Wes McKinney 4 , Weiwei Gong 3 , Xiangyao Yu 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-04701.txt.
Description
Bobbi Yogatama, Yifei Yang, and colleagues argue that analytical databases should be rebuilt around GPUs, and present Sirius, a prototype GPU-native SQL engine. They contend that recent hardware advances, including larger GPU memory, faster interconnects, and declining cost, have removed the barriers that limited GPU data analytics. Software has matured too, with composable data systems and solid libraries.
Sirius treats the GPU as the primary engine and uses libraries such as libcudf for high-performance relational operators. It offers drop-in acceleration for existing databases through the standard Substrait query representation, replacing the CPU engine without changing the user-facing interface. Integrated with the DuckDB system, Sirius delivered several-fold better cost efficiency on standard analytical benchmarks. Integrated with the distributed Apache Doris engine, it delivered speedups larger still. The argument: CPU-first database architecture is a legacy choice, and the tooling now exists to invert it.
Catalog record
| CREATOR | Bobbi Yogatama 1 ∗ , Yifei Yang 1 ∗ , Kevin Kristensen 1 , Devesh Sarda 1 , Abigale Kim 1 , Adrian Cockcroft 5 , Yu Teng, Joshua Patterson 2 , Gregory Kimball 2 , Wes McKinney 4 , Weiwei Gong 3 , Xiangyao Yu 1 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 30 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-04701