FSZ: Breaking the Prediction-Throughput Trade-off in GPU Lossy Compression
Distributed, Parallel, and Cluster ComputingPerformance
Summary
The authors developed FSZ, a new GPU-based data compressor that improves both how much data it can shrink and how fast it works. They redesigned the prediction step by sharing information across data blocks, choosing the best prediction method for each small section, and cleverly testing multiple methods at once without extra reading. These improvements let FSZ compress data better than previous methods while also running faster. Their tests on a powerful NVIDIA GPU showed FSZ beats existing top compressors in both compression ratio and speed.
Authors
Jiajun Huang
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.