Compression level 3
Benchmark index · Median overview · ← Level 2 · Level 4 →
Recorded run: long-20261109 (2026-11-09). Raw results · Environment · Run analysis.
Median across all datasets
Each of the 8 datasets has equal weight, including empty and tiny input. Speed / C = C mean latency / implementation mean latency; output / C = implementation bytes / C bytes. We take the median of these 8 per-dataset ratios separately for speed and size. 1× matches C; higher speed and lower output are better. These are medians across datasets, not median sample latencies, and do not imply statistical significance.
| Implementation | Median speed / C ↑ | Median output / C ↓ | Datasets |
|---|---|---|---|
| zstd C | 1.000× | 1.000× | 8 |
| mzstd | 1.268× | 1.000× | 8 |
Measurement setup and interpretation
zstd 1.5.7 (the C library built by zstd-ffi) and mzstd at the recorded
working tree compress identical bytes at this level. Every frame was checked byte-identical
to zstd’s and decoded back to its input before any timing.
These are cold native API measurements: construction, allocation, compression, and disposal. Complete known input; 50 samples, 3 s warmup, at least 6 s measurement. Host: AMD Ryzen 9 7950X, Linux KVM guest with two vCPUs: the encoder suite, then the decoder suite, each in one Criterion process pinned to vCPU 1; nothing else ran in the guest. Other tenants loaded the host: a fixed C decode varied by 8-43% (p10-p90 wall time) through the day, so sampling was lengthened to average over the noise. unset; Cargo release defaults (bench profile), runtime SIMD dispatch (AVX-512 level on this host; the decoder runs its block loops at the AVX2 level).
Chart whiskers and table intervals describe 95% confidence bounds on mean timing. They do not capture all host or allocator variation. Throughput bounds are transformed latency bounds. Output / input is compressed bytes divided by input bytes; smaller is better, and values above 100% mean expansion. For empty input, throughput and output / input are undefined (—). See the run analysis for measurement limitations.
Dataset details
Ordered by mzstd speed / C, highest first; all 8 datasets are shown. This order uses recorded mean latency, not statistical significance. Bars start at zero.
empty · tiny-text · repeated-1m · binary-64k · text-1m · random-64k · random-1m · alice29
empty
Empty input; measures API overhead. Input: 0 bytes.
mzstd speed / C: 10.353×; output: 9 / 9 bytes.
Exact measurements and confidence bounds
| Implementation | Mean, ns | 95% mean interval, ns | MiB/s | Output bytes | Output / input | Speed / C |
|---|---|---|---|---|---|---|
| zstd C | 2,207 | 2,163–2,260 | — | 9 | — | 1.000× |
| mzstd | 213.2 | 206–221.5 | — | 9 | — | 10.353× |
tiny-text
44-byte quick-brown-fox sentence. Input: 44 bytes.
mzstd speed / C: 6.912×; output: 53 / 53 bytes.
Exact measurements and confidence bounds
| Implementation | Mean, µs | 95% mean interval, µs | MiB/s | Output bytes | Output / input | Speed / C |
|---|---|---|---|---|---|---|
| zstd C | 2.545 | 2.467–2.628 | 16.49 | 53 | 120.455% | 1.000× |
| mzstd | 0.3682 | 0.3582–0.3806 | 113.98 | 53 | 120.455% | 6.912× |
repeated-1m
1 MiB of repeated a bytes. Input: 1,048,576 bytes.
mzstd speed / C: 3.340×; output: 50 / 50 bytes.
Exact measurements and confidence bounds
| Implementation | Mean, µs | 95% mean interval, µs | MiB/s | Output bytes | Output / input | Speed / C |
|---|---|---|---|---|---|---|
| zstd C | 144.8 | 141.5–149.4 | 6,904.64 | 50 | 0.005% | 1.000× |
| mzstd | 43.36 | 42.57–44.25 | 23,063.42 | 50 | 0.005% | 3.340× |
binary-64k
64 KiB of structured binary bytes: ((i / 16) XOR i) modulo 256. Input: 65,536 bytes.
mzstd speed / C: 1.391×; output: 2,324 / 2,324 bytes.
Exact measurements and confidence bounds
| Implementation | Mean, µs | 95% mean interval, µs | MiB/s | Output bytes | Output / input | Speed / C |
|---|---|---|---|---|---|---|
| zstd C | 17.49 | 16.89–18.15 | 3,573.62 | 2,324 | 3.546% | 1.000× |
| mzstd | 12.58 | 11.91–13.26 | 4,969.85 | 2,324 | 3.546% | 1.391× |
text-1m
Alice text repeated cyclically to 1 MiB. Input: 1,048,576 bytes.
mzstd speed / C: 1.145×; output: 55,878 / 55,878 bytes.
Exact measurements and confidence bounds
| Implementation | Mean, µs | 95% mean interval, µs | MiB/s | Output bytes | Output / input | Speed / C |
|---|---|---|---|---|---|---|
| zstd C | 720.3 | 696.4–744.9 | 1,388.34 | 55,878 | 5.329% | 1.000× |
| mzstd | 629 | 607.6–651.1 | 1,589.79 | 55,878 | 5.329% | 1.145× |
random-64k
64 KiB prefix of the deterministic pseudorandom corpus. Input: 65,536 bytes.
mzstd speed / C: 1.143×; output: 65,546 / 65,546 bytes.
Exact measurements and confidence bounds
| Implementation | Mean, µs | 95% mean interval, µs | MiB/s | Output bytes | Output / input | Speed / C |
|---|---|---|---|---|---|---|
| zstd C | 14.16 | 13.75–14.62 | 4,412.51 | 65,546 | 100.015% | 1.000× |
| mzstd | 12.39 | 12.03–12.78 | 5,045.02 | 65,546 | 100.015% | 1.143× |
random-1m
1 MiB of deterministic xorshift64 pseudorandom bytes. Input: 1,048,576 bytes.
mzstd speed / C: 1.020×; output: 1,048,609 / 1,048,609 bytes.
Exact measurements and confidence bounds
| Implementation | Mean, µs | 95% mean interval, µs | MiB/s | Output bytes | Output / input | Speed / C |
|---|---|---|---|---|---|---|
| zstd C | 126.1 | 122.3–130.1 | 7,928.03 | 1,048,609 | 100.003% | 1.000× |
| mzstd | 123.6 | 120.8–126.8 | 8,088.23 | 1,048,609 | 100.003% | 1.020× |
alice29
Alice in Wonderland text from the Canterbury corpus (testdata/alice29.txt). Input: 152,089 bytes.
mzstd speed / C: 1.011×; output: 56,970 / 56,970 bytes.
Exact measurements and confidence bounds
| Implementation | Mean, µs | 95% mean interval, µs | MiB/s | Output bytes | Output / input | Speed / C |
|---|---|---|---|---|---|---|
| zstd C | 565.5 | 551.9–580.3 | 256.50 | 56,970 | 37.458% | 1.000× |
| mzstd | 559.2 | 535.9–584.5 | 259.38 | 56,970 | 37.458% | 1.011× |