The compression libraries in this benchmark are rather unknown but represents the current top performers with respect to compression- and decompression speed.

The summary on the front page is the average of the values below, except for NotTheMusic.mp4 which is incompressible and special handled by QuickLZ which would make the bechmark unfair.

Except for QuickLZ C#, QuickLZ Java and zlib, this stand-alone benchmark program has been used:

    bench.zip (last edited 26-Sep-2009)

If you want to try it yourself, simply run bench_win32.bat or bench_win64 on Windows or run 'make gcc' in the C directory on *nix (run the resulting executables with a file name as argument).

Many CPU architectures and operating systems are supported by this benchmark program.

In the benchmark below all timings start directly before function call and end immediately after function exit and is therefore excluding disk I/O and operating system calls. They are executed in REALTIME_PRIORITY_CLASS using just a single core/thread and with multiple passes to eliminate timer resolution errors. All libraries are compiled for 64-bit x64 using Visual C++ 2008 with the /Ox flag and without initial performance profiling and executed on an Intel Core i7 920.

QuickLZ Sharp was executed from Visual Studio 2005 with Ctrl+F5 to enable JIT optimization. QuickLZ Java was compiled and executed with Java 6 SE, just using javac CMain.java; java CMain from the command line. No use of profilers or special compiler options.

Note that zlib and LZO have levels with better compression ratio and possibly faster decompression speed which have not been benchmarked.

Also note that performance differ on 32-bit x86 and RISC architectures. Simply run bench.zip above on your own system.