There is available CAL++ version of pyrit.
It's about 5% faster than OpenCL version on 4xxx cards.
On 5xxx it uses bitalign instruction and is ~40% faster than cpyrit-opencl.
Also it works on all ATI cards and only requires driver to run (no SDK).
I have a favour to ask. Maybe someone with 5870 card could post here benchmark results ( pyrit benchmark ).
Original pyrit project is located at http://code.google.com/p/pyrit/