3 weeks ago I was under the impression that as a current Mac Pro user I had around 2 options if I wanted to buy a new graphics card, which became 1 option if I wanted CUDA. Then I started reading more about it and found out that since Mountain Lion OSX 10.8, the vast majority of Nvidia cards and their CUDA power now run natively. Plug and play. Here is a list of supported cards.
2 weeks ago I bought a Nvidia GTX 660, plugged it straight into my 2008 Mac pro (running Mountain Lion) and everything worked perfectly.
1 week ago I bought a second 660, plugged it into my second PCI slot, and everything works perfectly (and twice as fast.)
Here's why I ended up with two new cards... I'm still unsure if its just better to have 1 really fast card, but right now I'm incredibly happy with how my system is running while using Octane.
Per Otoy, It is advised to keep a second card free for running the OS and C4D while using Octane. Because it utilises every bit of juice you allow it to on the card(s) you choose in the settings it causes the rest of the system to freeze up. In fantasy land I would love to be able to assign things on a molecular level but currently Mac OS apps 'choose' their GPU based on whichever monitor they launch on (therefore the GPU that is plugged into).
At home I have a slightly weird setup, my Mac Pro is running to 3 'monitors': a Cintiq 22HD (main monitor), a cheap Acer monitor, and my TV. The reason my Cintiq is my main monitor is purely for ergonomic reasons, I have a really small desk with room to only sit in one space - so thats where the Cintiq has to go, therefore its where my head has to mainly face.
Before buying the second 660, I tried sticking the (very) old Nvidia 8800 back into slot 2 and running my apps off that, leaving the 660 free to render.
So upon testing, if I plugged my main monitor (my Cintiq) into the old card, then C4D opens using that as its OpenGL processor - I can move around C4D as if nothing was even rendering, preferences show that openGL sees only the old card, which is great. But the huge down side to this is every application that launches on this monitor (all of them) will only ever use the old card unless you can specifiy otherwise (which you can't). This means once I progress beyond simple studio tests, my C4D viewport is slow again with a 5 year old card. Frustratingly, the only place you ever get to choose which card something uses is inside of Octane’s settings!
In a practical sense, to get by, if you want to use the powerful card for C4D and all your other programs (and why wouldn’t you) then there is a simple checkbox that forces a low render priority (similar to the ‘thread priority’ in Vray, how it deals with the CPUs).
Make sure to choose ‘Priority Low’ in the dropdown and check the ‘use priority’ box next to the actual card. Also tell it not to use anything you don’t want it to. My old card did not help render speed, in fact it slows everything right down because the scene has to load into all the GPUs before rendering so it sucked all the speed form the 660.
So my current setup is 2xGTX660, the card plugged into PCI 1 is running all my apps at a blazing speed while set to 'Low Priority' is giving Octane as much as it can, then the second card is giving Octane everything is has! This literally halves render time again. I can't recommend that this is the best way to go, I can only document my experience, which so far has been amazing.