ActuaLLY, on a nvidia card, cpu activity is 3%. ATI is the hog. But using enviromental variables, you can knock the cpu down substantially.
Unicores, or smp, is really up to you. As both work are needed, however, due to the shorter deadlines, and larger wus, smp usually nets more ppd.
I use to run smp on 2 cores in linux vmware, and net as much as 4 cores in windows... but my comp hated to run smp and locks up after 2 days.
My dual core smp nets 500+ppd, while my dual unicore laptop is like 400 if i run both.