May 12, 2026
Compressed files, uncompressed egos
Fc, a lossless compressor for floating-point streams
This tiny data-squeezer wowed number nerds, but the comments turned into a showdown
TLDR: fc is a new tool for shrinking huge streams of numbers, and on its own tests it often makes smaller files than better-known options. Commenters were impressed but instantly turned it into a benchmark battle, demanding comparisons and debating whether its many-tricks approach is genius or overkill.
A new project called fc landed with a very specific promise: make long streams of decimal numbers take up less space without losing a single bit. In plain English, it’s built for giant logs of measurements, prices, or science data, and on its own tests it often beats big-name tools when the goal is the smallest file possible. That alone would be nerd catnip. But the real action? The comment section immediately turned into a live audition for every rival idea in the room.
Creator enduku tried to set expectations early — this is not coming for zstd or lz4’s crown, it’s a specialist, not a universal champ. Naturally, the community responded by doing exactly what communities do: comparing it to absolutely everything. One commenter name-dropped Chimp128 and Arrow’s “byte stream split” like they were surprise wrestlers entering the ring. Another demanded a face-off with pcodec. Someone else tossed in OpenLZ, basically asking, “Cute benchmark, but can the grown-ups do this too?”
The funniest mini-drama came from the “metacompressor” label. userbinator clocked fc’s whole vibe as a machine that makes many compression tricks fight each other and picks the winner — which sounds smart, but also triggered the classic hot take: does one method secretly win all the time anyway? That’s the mood in a nutshell: impressed, skeptical, and deeply unable to resist turning a promising release into a comment-thread cage match.
Key Points
- •fc is a lossless compressor for IEEE-754 64-bit double streams that uses adaptive blocks and per-block competition among specialized codecs.
- •The article reports aggregate benchmark results across 17 synthetic datasets, where fc achieved a 3.07 compression ratio, 120 MB/s encode throughput, and 1277 MB/s decode throughput.
- •In the published benchmark, fc achieved the best compression ratio on 10 of 17 datasets and was described as the best size-oriented floating-point compressor in that test set.
- •fc performs best on structured, analytic, periodic, and quasi-periodic floating-point data, while zstd performs better on certain quantized datasets and fpzip is slightly better on several noisy float arrays.
- •The library is described as research-grade, lossless only, tuned for double-precision floating-point patterns, and limited to x86-64 systems with AVX2, SSE4.2, BMI, and LZCNT support.