OK so lets look at your render needs.
About how many polys do your scenes usually have?
Do you make extensive use of textures?
Do you need to render high resolution images?
All of these? - because these will determine how much vram you are going to need and therefore lead you to narrow your choice.
Further you need to ask yourself how much you want to spend or if you mind waiting a little longer for a render or not.
If you need a chunk of vram there are only a few choices in gaming cards and then you need to move up to Quadros and Teslas which dont really perform any different for the same number of cuda cores but cost a whole lot more.. They range 2,2.5,3,4 and 6gb
There are some 2xx series cards that came in 1792mb and 2gb variants but these are surpassed by the Fermi series in performance.
Assuming you need the vram as a priority the 460/2gb are hard to pass up at this time.
They offer decent performance, capacity and are quieter and cooler than the 465,470,480.
As mentioned previously the 460 may benefit from cuda3.2 in that it might allow more cuda cores to be utilised than present ie 224 out of 336.
The 460 being half a 480 performance wise is based on the 224 cores. By the same measure a 460 is about 2/3 of a 470.
If you work in close proximity to your render pc you might appreciate the physical virtues of the 460 and the cards might have a longer life in constant heavy use.
If you arent so concerned about vram but want to render faster as a priority then the 470 is attractive I think however if the 460 benefit from cuda 3.2 that would be in question.
You should wait a little while longer before purchase to see if that is going to occur or not.
There is rumoured to be a 475 in the pipeline (384 cores) that might be generally comparable to a 470 but with the virtues of the 460 but I think it will be a 1gb card.
You need to think too about what the speed difference means in practice. If you dont mind waiting 15 mins instead of 10 the case for extra $ etc isnt compelling. If you have a render to do urgently for a deadline it might be worth doing.
If you are going to use 2 cards for Octane rendering and buying anew I would get a third basic card for the UI display however being a basic card may not make it so suitable to other tasks.
So you need to ask yourself what you want to do in the case of other applications.
The most obvious choice is to run one or both 460/470 with a monitor attached and let the basic or other card idle at these times and swap into render configuration as you need.
Another option is to use a Quadro580 or 1800 for regular use + professional apps + as the Octane UI card, along with the 2x460/470.
These cards are fairly capable but not too expensive or power hungry.
There might be a couple of Fermi based replacements for these soon, not sure.
You will need to mod the Quadro drivers and install manually to allow the 460/470 to run on the same one.
This is a bit of a hassle to do and the Quadro drivers are slightly behind in the update cycle.
Looking forward there is a vague possibility you might not be able to share a driver between a Quadro and a Geforce like this. Who knows how things will be come cuda5.0 and how Octane will develop and what driver it might require then. It would pay to pick the third card to be from the same generation as the render cards if possible for these driver compatibility reasons.
I would go with a 850W PS I think just for a little headroom. Have a read of the forums and ask opinions of other users too. I am offering some assistance but I dont know everything thats for sure.
