December 20, 2025
Compression beef, decompressed
Show HN: ZXC – Asymmetric, +40% decode vs. LZ4 on ARM (C, BSD-3, Fuzzed)
40% faster on Apple chips? Devs cheer, skeptics clap back
TLDR: Open-source ZXC promises much faster file unpacking, especially on Apple chips, by accepting slower packing upfront. Commenters love the independent [lzbench] tests but argue the pitch mislabels LZ4, sparking debate over marketing vs. reality and whether ARM gains matter more than modest x86 improvements.
ZXC just showed up claiming it makes apps and games load faster by unpacking files up to 40% faster on Apple chips and 22% faster on cloud ARM servers. The twist: it packs files slower on purpose, so they can be unpacked super fast later — perfect for “write once, read many” releases. Devs swooned at the idea of snappier startup, while others side-eyed the marketing. If you compress assets once during build and unzip them millions of times on user devices, ZXC sounds like a win. But that’s where the comment war began.
One camp waved receipts: ZXC is in lzbench, an industry test suite, so “apples to apples” comparisons are on. The other camp pounced on the pitch: “Unlike LZ4” — wait, LZ4 already does the slow-pack/fast-unpack thing, a commenter snapped. Cue memes about “marketing math,” ARM fans, and x86 dads: “only +4%.” Joked CI engineers will build a shrine to patience as users get speed. Others asked for multi-thread and real-world asset tests. One link-dropper pointed to the repo so anyone can check the numbers.
Key Points
- •ZXC is an asymmetric lossless compression library designed to maximize decompression throughput for content delivery and embedded systems.
- •Verified lzbench integration allows independent validation; continuous single‑threaded benchmarks run via GitHub Actions on ARM64 and x86_64.
- •On Apple Silicon (M2/M3), ZXC -3 decodes at 6,365 MB/s (1.39x faster than LZ4) with slightly smaller size (-1.6%); ZXC -5 is 3.3x faster than Zstd‑1.
- •On Google Axion (ARM Neoverse V2), ZXC -3 achieves 5,084 MB/s (1.22x faster than LZ4); on AMD EPYC x86_64, ZXC -3 is +4% faster than LZ4.
- •Benchmarks used Clang 17.0.0 on macOS Sequoia and GCC 12.2.0 on Debian, with the Silesia Corpus and -march=native compilation.