Let me just throw in my two cents on all this since I got tagged.
First off, as some have mentioned "compute" cards, that brings to mind things like the Intel Phy, the NVidia P100 and others. They cannot be used for F@H or mining. Those are specifically designed cards for running custom software, usually a cluster based solution, used for rendering and massive compute solutions. Not something you can run Windows on nor do they even make a F@H client that will recognize those cards.
Now, you have to look at the F@H client itself before going nuts on video cards. The folding on Linux builds is still iffy and a lot of work, so I won't even go there. When it comes to Windows, getting more than 5 GPU's supported and running correctly in Windows is pretty much impossible. It can be done, but for the work it isn't really worth it. So, since you need to run Windows you are pretty much looking at a max of 5 GPU's for a stable system. Sure, you could buy 16... and that won't work at all. And again, you need to look at your motherboards and such as many of them have limitations on the number of PCIe slots active because of electrical wiring. All in all, getting a 5 GPU solution to work is very doable if you do your homework. More than 5 GPU's and you are looking at probably a lot of issues.
The recommendation to leave at least one core per GPU free was mostly for NVidia, due to the way their drivers work, but was also good advice for AMD cards as well. The client when initially setup would original default to use all CPU cores (hence the issue with NVidia cards) and then was changed to use all but one. But then if you had two GPU's you still needed to modify it and... some projects didn't like an odd number of CPU cores, blah blah... Overall, for the heat it generates and the extra cooling you need, don't even bother folding on the CPU. Just fold on the GPU(s) and you will pretty much maximize you solution. If you have a monster CPU you don't want to go to waste, just do "light" folding with the Chrome NACL client and voila.