arXiv:2609.06635 · v2
From JPEG to CUDA Batch: Measuring RGB Input Pipeline Throughput
摘要
We measure how many ready batches per second eight image input pipelines deliver to an NVIDIA L4. Each pipeline starts with RGB JPEG files and delivers synchronized CUDA float16 batches of 256 images. The benchmark contains 1314 measurements across 73 recipes. On the 34 recipes shared by all eight paths, DALI delivered a median of 19.7 batches/s and AlbumentationsX (AX) delivered 17.1. DALI was fastest on 34 of those recipes. Median peak process GPU memory was 1,872 MiB for DALI and 404 MiB for AX.
作者备注:31 pages, 6 figures. Expanded the benchmark to 73 recipes and eight input paths, revised the comparisons, and updated the results appendix. Code and data: https://github.com/albumentations-team/benchmark/tree/154baa0da1880013941daad1b6bd35c58097f969