Love it so much.
Using: -k phatk DEVICE=0 VECTORS BFI_INT FASTLOOP=false WORKSIZE=128 AGGRESSION=12 on a single 5870 950/330 I get a nice 410 Mhas/sec.
phatk works best with
WORKSIZE=256. Getting 413 Mhash/s on both 5870 935/319 in a dual-card mining rig (w/o CrossFireX bridges).