I use WinAFC 0.92 with a fair bit of success.
There are two approaches to this whole GPU/CPU co-existence scenario:
1) Allow CPU affinity changes by running the SMP client with the -configonly flag and selecting "yes" when it asks to allow CPU affinity changes
2) Use an affinity changer to lock all GPU clients to one core, and preferably a PHYSICAL core with Core I7 (920/980x)
3) Run the -smp (n-1) flag with n being number of cores/threads, so -smp 11 for Core I7 980x or -smp 7 for Core I7 920/860 and lock all GPU clients to the unused (by SMP) core either physical or HT.
Right now my Core I7 980x @ 4.05 Ghz is getting over 50K on a P2685 with three GPU clients running using the WinAFC setup to lock all GPU clients to a physical core.
But yes, I notice my Phenom II X6 or my Opteron (24-core) duallie box see less slow down with multi, or single GPU clients running simultaneously with SMP. I guess real cores seem to make some differences.
|