FaCTz: GPU Compressor Preserves Vector-Field Topology at 60 GB/s

Error-bounded lossy compression is critical for storing and transferring vector-field data from large-scale scientific simulations. Existing GPU compressors achieve high throughput but ignore the field's topology, meaning small allowable perturbations can silently create or eliminate the critical points that downstream feature analysis relies on. The only compressor that guarantees critical-point preservation, cpSZ, runs on the CPU and falls far short of the data rates modern GPU systems generate.

FaCTz is presented as the first GPU-based error-bounded lossy compressor that guarantees critical-point preservation. The key insight is that although preserving critical points is inherently a sequential, coupled constraint, it can be restructured into independent parallel tasks, either per-block or, speculatively, per-point. FaCTz offers a block-wise mode tuned for throughput and a speculative per-point mode tuned for compression ratio.

Across three vector-field datasets, FaCTz preserves every critical point while reaching throughput of up to 60 GB/s, roughly 640x faster than the multithreaded CPU implementation of cpSZ. Its speculative mode further boosts compression ratio by about 2x over the throughput-oriented mode, giving engineers a practical way to combine GPU-scale speed with guaranteed topological correctness in simulation pipelines.