Publication
Do GPUs Really Need New Tabular File Formats?
Jigao Luo; Qi Chen; Carsten Binnig
In: Proceedings of the 22nd International Workshop on Data Management on New Hardware, DaMoN 2026, Bengaluru, India, 31 May 2026- 5 June 2026. International Workshop on Data Management on New Hardware (DaMoN), Pages 9:1-9:6, ACM, 2026.
Abstract
Apache Parquet is the de facto columnar file format in modern
analytical systems, yet its configuration guidelines remain largely
shaped by CPU-centric execution engines. As GPU-accelerated data
processing becomes increasingly prevalent, Parquet files optimized
by default for CPU-oriented engines often severely underutilize
GPU hardware, unnecessarily turning GPU scans into an artificial
performance bottleneck.
In this work, we systematically study how Parquet configura-
tions affect GPU scan performance, and show that poor Parquet
performance on GPUs is mostly a consequence of suboptimal config-
uration choices. By applying GPU-aware configurations, we achieve
up to 125 GB/s effective read bandwidth using fully compliant Par-
quet files, without switching to a new file format or sacrificing the
interoperability benefits of the de facto standard.