diffpak

A differential compresor (or a binary diff) for huge files. Unlike other tools, e.g. xdelta3, it searches for matching data through the whole source file, even if it weights several gigabytes, while using much less memory (with default configuration approx 25x less than the size of the source file). Output files are not compressed, so you can use any compressor you like with great results.
It is quite fast for very similar files (about the speed of hdd, however it reads input files twice) and not much worse on files with lots of differences.