While I have an NVIDIA Geforce GTX 275 Mac Edition, it is very much old tech with only 240 CUDA cores for working with Premiere Pro. And it is about the fastest there is (other than a QUADRO 4000, but even that is old tech compared to current PC video card technology).
One solution is MacVidCards and their Ebay Site. These guys flash GTX 5xx series videos card with Apple Firmware so that you get full functionality on a Mac, including boot screens and EFI (so if you hold down option you can select hard drives, or go into the terminal before you hit the desktop). This is a great and easy solution for most people, though it leaves out Kepler video cards, the GTX 6xx series cards that are the latest and greatest with the most CUDA cores for real time playback in Premiere Pro.
As of Mountain Lion putting in a GTX 6xx series or Kepler has gotten a lot easier. This is because Mac OS X and the latest NVIDIA drivers now include drivers for the MacBook Pro 15″ Retina Display which is has Geforce GT 650m, so there is now some Kepler support in OS X. This means that PC Kepler cards will run in Mountain Lion, with a couple of caveats. One is you don’t get the boot screens, because the video drivers don’t kick in until just before the desktop shows up. The second is that due to a setting in OS X you can only have up to 2GB of RAM in a video card, if you have more you won’t get OpenCL acceleration. It is on this point that the excellent Netkas site comes in. Netkas has figured out multiple methods to make 6xx series NVIDIA cards with over 2GB of RAM work on a Mac.
Yes there are bigger cards with more CUDA cores, like the GTX 680, but it requires 8 pin power, and the Mac only supplies 6 pin power, so you will have to run an external power supply to get it to work, and the GTX 690 is basically 2 video cards in one, so the mac will have trouble using the dual cards, and you won’t get the benefit (except in boot camp into Windows). Basically the 670 uses 2 6 pin power adapters, so your mac can power it. I picked up an EVGA Geforce GTX670 FTW+ 4096MB GDDR5 Graphics card and installed it.
There are the instructions for a GTX 670 and the instructions for the rest of the 6xxx series cards on how to get OpenCL working on a card with more than 2GB of RAM.
Now I had problems getting the GTX 670 instructions to work.
I started to a TechTool Pro Safety partition I have on a second drive and did the replacement and repaired permissions and OpenCL was not fixed. After much fiddling I couldn’t figure it out, and having read elsewhere on the Netkas forums to also do the Hex edit, I also did the 32 and the ML Hex edit fixes, repaired permissions (A VERY IMPORTANT STEP AS YOUR COMPUTER MAY NOT BOOT IF YOU DON’T DO IT). I used the excellent HexEdit to edit the file.
I tried getting help on the forums, but was basically told how stupid I was and that I had done it wrong (which was not the case). Finally I decided to re-install the latest NVIDIA drivers and see if that did anything, and after restarting, I ran LuxMark (to test OpenCL speeds) and low and behold it now worked.
Honestly I am not sure if I just needed to do the Hex Edit only, as the instructions I saw at Netkas said to try the file replacement for the 670, or if it really did require me doing both to get OpenCL acceleration activated, but doing both certainly worked for me with the addition of re-installing the NVDIA Drivers.
The next step is now to enable CUDA for your new video card in Premiere Pro and After Effects CS6. VidMuze has an excellent video on how to do this, and a download file that gives specific instructions. This was simple in comparison to the hex file thing, which was not hard, but certainly not this easy, but this does require some terminal use, so you should be comfortable using the terminal in your mac.
I now have 1344 CUDA cores for Premiere Pro and After Effects, and am looking forward to taking Premiere Pro CS6 and the Mercury Playback engine for a spin with more than 5 times the CUDA cores, twice the RAM and a much faster and newer processor. I will let you know how it performs.