Dual GPU cards only use a single GPU with PR. So the 590/690 are only used for 50%. The 580/680 are single GPU cards.
Video card performance
With the number of observations available now (980+) we see a clear tendency that more CUDA cores do help performance, at least with the older architecture. The usual performance increase of hardware acceleration over software is around 12 x, with the cards with fewer CUDA cores lagging behind and the ones with more cores pulling ahead. The increase in performance over software shows that indeed the 9800 GT and GTX 260 are clearly lagging when compared to the GTX 480/580/680. The newer architecture shows its benefits.
On the other hand, with the introduction of Kepler cards with around three times the number of CUDA cores over the Fermi cards, people expected triple the performance and they were disappointed to be shown wrong in their expectations.
With all the results we have from our PPBM5 benchmark it sure looks like memory bandwidth may be the decisive factor to impact performance. It is no longer just number of CUDA cores. Time will tell.
The much touted Maximus solution, at least by Adobe, is an utter waste of money, because it requires a very expensive Quadro card plus a Tesla C2075 card, that is even slower than a simple two generations old GTX 470, for the simple reason that it lacks CUDA cores and memory bandwidth to make it faster. We have several Maximus solutions in the current PPBM5 benchmark and despite prices up to € 6000 for the Quadro 6000 plus a Tesla C2075, these solutions are easily outperformed by many GTX 470/480/570/580/670 and 680 cards, that cost only a fraction. The only thing impressive about a Maximus solution is the price. The only reason to opt for a Maximus solution, despite the cost, is if you absolutely need 10 bit output to your very expensive 10 bit monitors, and then you only get quality, not performance.
What determines video card performance for PR?
Based on information currently available, it is not so much the number of shaders or CUDA cores. If that were the case, all Kepler cards would easily leave Fermi and older cards in the dust with about three times the number of CUDA cores, but that is not the case. The determining factor may be memory bandwidth and that explains why the Kepler range is only slightly faster than Fermi and why Kepler is significantly faster than all Quadro cards.
Do i need to upgrade to 680, i dont need more displays or more ram or cuda cores iam happy with GTX 580's?
The super short summary answer I give to people is the onbaord video RAM speed on the card has much more of an impact on peformance than the width of the data being fed in. Generally, the bigger bits, the more data it's feeding in at once but there's more math involved than that. I would bet Adobe natively supports whatever the graphics card does but even if it has to convert it on the fly, it's so good at it that it's a couple percent difference. So if you find an OCed card or one with exceptional GDDR5 that runs at 3900MHz instead of 3800MHz, it will blow away the performance compared to any slight memory operation size matching.