EBLC Benchmarking Toolkit
A standardized benchmarking and analysis toolkit for GPU-accelerated, error-bounded lossy compressors. Work in progress.
EBLC Benchmarking Toolkit (benchkit) is a benchmarking and analysis harness for GPU-accelerated, error-bounded lossy compressors (cuSZ, cuSZp, cuSZ-Hi, MANS, PFPL, and more), built to validate FZGPUModules ports of these compressors against their original implementations on compression ratio, throughput, quality, and memory.
The harness computes every quality and size metric itself from raw compressed/decompressed artifacts rather than trusting each tool’s self-reported numbers, normalizes error-bound semantics across tools before comparing them, and records full provenance (GPU, clocks, driver, commit SHAs, dataset checksums) so results are reproducible months later for a paper revision.
Still early and actively evolving — more details, figures, and a public repo link to come as the toolkit matures.