FSZ: Breaking the Prediction-Throughput Trade-off in GPU Lossy Compression

2026-07-16

AI Summary

This work addresses the challenge that existing GPU lossy compressors struggle to achieve high compression ratios under high throughput due to their reliance on fixed first-order prediction. To overcome this limitation, the authors redesign the prediction phase of the GPU core-level compression pipeline and introduce three synergistic innovations: cross-block prediction state propagation, tile-wise adaptive multi-order prediction with centralized policy selection, and a single-pass four-candidate evaluation mechanism. Built upon the CUDA core architecture and integrating Lorenzo prediction, finite-difference mathematical properties, and adaptive strategy selection, the proposed method achieves up to a 10.95× improvement in compression ratio on NVIDIA GH200 while simultaneously delivering 676 GB/s compression and 785 GB/s decompression throughput—significantly outperforming current state-of-the-art approaches.

Abstract

Existing fast GPU error-bounded lossy compressors have achieved high throughput through pure-GPU single-kernel designs, but their compression ratios remain limited because they typically apply a fixed first-order predictor on independent blocks. We propose FSZ, a GPU error-bounded lossy compressor that redesigns the prediction stage with three mutually reinforcing algorithmic innovations to achieve both higher compression ratios and higher throughput within a single CUDA kernel: (1) cross-block prediction state carries Lorenzo prediction state across block boundaries within 256-element tiles, eliminating 7 out of 8 boundary residuals that inflate encoding rates; (2) per-tile adaptive multi-order prediction and centering adaptively selects the best compression strategy per tile from first-order, second-order, and centering variants; and (3) a single-pass four-way evaluation exploits a mathematical property of finite differences to evaluate all variants from a single data read, enabling richer prediction within the same bandwidth budget as a fixed predictor. Experiments on NVIDIA GH200 GPU with 8 real-world application datasets show that FSZ outperforms cuSZp-P by up to 10.95x and the state-of-the-art cuSZp-O by up to 2.92x in compression ratio. Notably, these gains come with no throughput penalty: FSZ simultaneously achieves the highest average throughput (676 GB/s compression, 785 GB/s decompression) among all evaluated compressors.

Problem

• GPU lossy compression
• compression ratio
• throughput
• error-bounded compression
• prediction

Innovation

• cross-block prediction
• adaptive multi-order prediction
• single-pass evaluation
• GPU lossy compression
• error-bounded compression