Home Benchmark Docs License Get Early Access

v1.0.1 — Now Available

GPU where it matters.
CPU where it doesn't.

HammerIO automatically routes files to the fastest available hardware. Large files hit nvCOMP GPU LZ4 at 6+ GB/s decompression. On macOS, Apple LZFSE via the Accelerate framework. Everything else goes to CPU zstd with parallel threading. No configuration required.

8,537 MB/s GPU Decompress
4,258 MB/s Roundtrip Decompress
4.3x GPU vs CPU
Apache 2.0 License

How it works

Drop it

Right-click any file or folder. Select HammerIO: Compress. Done. No terminal required.

Routed automatically

The JobRouter profiles every file — size, type, entropy — and sends it to the optimal engine. GPU for large files. CPU for small ones. Passthrough for already-compressed data.

Verified

Every operation includes SHA-256 integrity verification. Compress, decompress, verify — all in one command. match: True or it failed.

Numbers that hold up

Measured on Jetson AGX Orin 64GB, JetPack 6.2.2, CUDA 12.6, MAXN mode. 10 GB roundtrip verified.

Engine Size Compress Decompress Ratio Integrity
GPU nvCOMP LZ4 10 GB 517 MB/s 4,258 MB/s 1.98x PASS
CPU zstd-1 10 GB 1,094 MB/s 733 MB/s 2.00x PASS
CPU zstd-3 10 GB 1,014 MB/s 741 MB/s 2.00x PASS

Full benchmark results →

Built for real workloads

Smart Routing

HammerIO profiles every file before compressing. Size, file type, and entropy determine the optimal engine. GPU overhead only applied when the speedup justifies it.

Large Files → nvCOMP LZ4 (GPU) # Files > ~10MB Datasets → nvCOMP LZ4 (GPU) # CSV, Parquet, NPY macOS → Apple LZFSE # M1/M2/M3/M4 General → zstd parallel (CPU) # Default path Archives → passthrough # Already compressed
FileGPUCPU

Native Integration

Right-click integration for Nautilus, Thunar, and Nemo. Desktop notifications with before/after size. Works like any other app.

Right-click → HammerIO: Compress

Watch Daemon

Drop files into compress/ — they get compressed. Drop .lz4 or .zst files into decompress/ — they get decompressed. Originals moved to processed/. Fully automated. No supervision required.

pipeline/ compress/ ← drop files here decompress/ ← drop .lz4/.zst here compressed/ → output here decompressed/ → output here processed/ → originals archived

Hardware Compatibility

Runs on any NVIDIA GPU with CUDA support, Apple Silicon via LZFSE, and automatic CPU fallback when no GPU is present.

PlatformEngineStatus
Jetson AGX Orin 64GBnvCOMP GPU LZ4Primary — fully tested
Jetson AGX Orin 32GBnvCOMP GPU LZ4Supported
Jetson Orin NX/NanonvCOMP GPU LZ4Supported
RTX 3000/4000/5000nvCOMP GPU LZ4Supported
Any CUDA 12.x GPUnvCOMP GPU LZ4Supported
macOS Apple SiliconApple LZFSEM1/M2/M3/M4
CPU-onlyzstd / gzip / lz4Universal fallback

Get Early Access

HammerIO open source release coming soon. Commercial licenses available now.

Already need a commercial license? Contact us View license terms →