Graig,
In general more CUDA cores is better, but with the advent of Kepler, people thought they would get around three times the performance of a 580, because the number of CUDA cores had tripled. Not so. I believe that it is all about the memory bandwidth that determines the performance of a video card. That is why the 570/580/670/680 are clearly faster than a Quadro 4000/5000.
Modern day cards have a lot of VRAM and for most users that is enough for daily work, but if you have used all available VRAM, PR automatically and ungracefully reverts to software MPE mode. If you only edit HD material, you will be hard pressed to create a situation where you run out of VRAM, but if you edit Red 4K, Epic 5K and similar high resolutions, the situation of running out of VRAM can occur much more often.
If you want to buy a card now, look at the best memory bandwidth you can get and the maximum memory you can get. Then consider your options.
Example:
- GTX 680/4 GB - memory bandwidth 192.3 GB/s, 4 GB VRAM, price around € 600.
- Quadro 6000 - memory bandwidth 144 GB/s, 6 GB VRAM, price around € 3600.
The GTX 680 clearly delivers better performance, has quite a lot of memory for around 1/6-th the price of the Quadro.
