The two main clients in the F@H project are the SMP [Symmetric Multi-Processing] client [CPU client] and GPU client.
A GPU will generally give much higher PPD than
most CPUs running the SMP client. I run two GTX260s and they give me 8000 points per day each. My Q6600 gets 4K on a good day. However, GPUs use way more juice and put out a lot more heat than any CPU client ever would. So it is balanced. You can get maximum output, but at the cost of higher leccy bills, or lower output and lower leccy bills.
However, with i7s becoming more and more popular GPU's aren't the kings of output anymore. My i7 920 at 4GHz chucks out about 17K PPD if I don't do much with my computer. I need
two GTX260s just to get 16K PPD. I don't think any GPU can push out that much PPD on its own [corrections welcome and dual-core GPUs don't count as they're just two lower spec GPUs nailed to one PCB!

]. Even if they did, they will no doubt burn a lot of juice to do that! Plus you are looking at the £250+ price bracket for one too.
If you have an i7 machine you barely use for anything other than browsing the Internet, you could also run the SMP client with a special option to do really huge Work Units. These take 2-3 days to complete but come with a massive 50,000 point bonus. This works out at around 23-25K PPD. Though this requires massive amounts of RAM and isn't for casual folding!