http://clbenchmark.com/compare.jsp?config_0=11972906&config_1=11954866
Reviews from launch in March are NOT good places to find numbers. They were on early BETA drivers.
This is all in Windows, keep in mind.
Nvidia (like APple and everyone else) runs a business to make money.
They make FAR MORE PER CARD on Tesla/Quadro type cards.
They don't actually develop 2 different GPUs, they develop one and then figure out ways to make the PRO cards work better at PRO apps.
With Fermi they made the "Pro" cards have 4 times the Double Precision Float abilities. Since no Mac app that I know of uses DPF, this means that a GTX card with similar specs will slaughter a Quadro/Tesla card in performance per dollar in EVERY SINGLE APP CURRENTLY OUT THERE. The only other special thing about Quadro Tesla is amount of RAM. The QUadro 6000 has 6 GB, so there might be an app somewhere that can use this. If someone knows of one, please post it.
Point being, it is very hard for ANYONE to justify spending $$$$ on a Quadro when a GTX for $$$ will leave it for dead in rendering in OSX. (You read it right, you will pay MORE for LOWER performance in most cases if you chose Quadro/TEsla)
Nvidia isn't a bunch of idiots, they realize that plenty of people will pony up the BIG BUCKS for a card if it can be shown to increase their output. So Nvidia has hobbled GPGPU on Kepler a new and better way. The consumer cards are designed to EXCEL at gaming type calculations, leaving the GPGPU mastery to the (yet to be released) Pro level cards which will likely give CUDA & OpenCl scores the likes of which we have never seen. They will also cost $$$$.
Right now this means GTX5xx cards are quite a deal since they offer un-hobbled GPGPU performance. The GTX6xx have been partially castrated.