Incidentally I have 2 x 7950 and a crappy quad core with code I have compiled myself. I have also tweaked a few things too (but nothing like the HP10 improvements) and have minded 2 blocks with it in about 20 hours.
OK good,
some schreenshot of that phenomen ?
how did you setup, GPU doing the test, or ?
protocol primecoin
worksize 256
aggression 21
sharethreads 1
sievepercentage 10
sievesize 2000000
platform 0
device 0
device 1
vectors 2
lookup_gap 2
debug false
opencldebug false
cpu_mining_threads 8
use_gpu true
use_gpu_fermat_test false
getwork_rate 4000
gpu_thread_concurrency 24000
Like I said I have made a few tweaks to the code which might give me 5% increase but nothing much. The result is....
61152.9 fermats/s, 3581.43 gandalfs/s.
354.966M TOTAL 1.02913M 2-chains 65.0499k 3-chains 4.18879k 4-chains 286.323 5-chains 16.1493 6-chains
... which appears to be nothing special really. I think I'm very lucky to have found 2 blocks already
edit: this is using ubuntu.