CPX Performance Benchmarks
January 28, 2026 · View on GitHub
Methodology
- Benchmarking Tool: hyperfine
- Runs per Test: 6 iterations with (cache cleared before each test)
- CPX Configuration:
-r -j=16(16 parallel threads) - GNU cp Configuration:
-r(standard recursive) - Dataset: 11 major open-source repositories (773,996 files, 13 GB)
- Hardware: Intel Core i7-1165G7 (4C/8T), SSD storage
- OS: Ubuntu 25.10 (Kernel 6.x)
- Consistency: Results validated across multiple Intel/AMD platforms with <5% variance
Overall Results
Full Dataset (13 GB, 773,996 files)
cp:
| Tool | Mean Time | Speedup |
|---|---|---|
| cpx | 28.72s ± 1.46s | 2.81× faster |
| cp | 80.56s ± 2.79s | baseline |
Time Saved: 51.84 seconds
rsync:
| Tool | Mean Time | Speedup |
|---|---|---|
| cpx | 30.37s ± 0.82s | 2.67× faster |
| rsync | 80.54s ± 2.567s | baseline |
Time Saved: 50.54 seconds
Per-Repository Results
cp:
| Repository | Files | cpx | cp | Speedup |
|---|---|---|---|---|
| VSCode | ~15k | 263ms | 1,084ms | 4.12× |
| Rust | ~65k | 1,091ms | 4,553ms | 4.17× |
| Go | ~12k | 336ms | 1,373ms | 4.09× |
| TensorFlow | ~50k | 736ms | 2,846ms | 3.87× |
| Kubernetes | ~35k | 744ms | 2,805ms | 3.77× |
| Node.js | ~45k | 1,351ms | 4,573ms | 3.39× |
| Linux Kernel | ~72k | 2,925ms | 9,314ms | 3.18× |
| Chromium | ~450k | 15.89s | 46.19s | 2.91× |
| OpenImageIO | ~5k | 125ms | 258ms | 2.07× |
| Godot | ~8k | 354ms | 550ms | 1.55× |
| OpenEXR | ~2.5k | 273ms | 286ms | 1.04× |
rsync:
| Repository | Files | cpx | rsync | Speedup |
|---|---|---|---|---|
| Kubernetes | ~35k | 617.2ms | 3,097ms | 5.02× |
| Rust | ~65k | 1,014ms | 4,604ms | 4.54× |
| Go | ~12k | 340ms | 1,529ms | 4.49× |
| TensorFlow | ~50k | 736ms | 3,100ms | 4.21× |
| Node.js | ~45k | 1,198ms | 4,821ms | 4.02× |
| VSCode | ~15k | 310ms | 1,190ms | 3.83× |
| Linux Kernel | ~72k | 2,587ms | 8,868ms | 3.43× |
| Chromium | ~450k | 14.76s | 47.74s | 3.23× |
| OpenImageIO | ~5k | 120ms | 385ms | 3.21× |
| Godot | ~8k | 392ms | 987ms | 2.52× |
| OpenEXR | ~2.5k | 267ms | 588ms | 2.20× |
See all hyperfine reports(cp,rsync,xcp) here https://github.com/11happy/cpx/tree/main/benchmarks
Why faster ?
CPX advantages:
- Parallel I/O operations (16 threads ) [can be more depending on system]
- Adaptive buffer sizing (64KB-2MB)
- Parallel directory traversal
Reproduction
# Clone repository
git clone https://github.com/11happy/cpx
cd cpx
# Run cold cache benchmark (requires sudo), adjust the paths in bench.sh accordingly
sudo ./benchmarks/bench.sh cold
# Results saved to ~/cpx_multi_bench/
Note: Performance varies with storage type (NVMe > SSD > HDD), CPU cores, and filesystem (ext4/btrfs/xfs).