Hey there crazy
I have just got a GTX460/2gb. it replaced a GTS250/1gb.
I 've been trying it out for size and I think 2gb is very useful.
Its actually probably more space than I usually need cos my renders are industrial design stuff and machinery with so not many textures but I thought the extra space was too good to pass up for just a few more $. I anticipate for most stuff I do I will need less than 1gb - the 1gb I had wasnt too limiting.
I also needed something which will fit in my pc case and run off my 550W PS.
I was aware that Octane rendering for hours is pretty hard on the card temp wise and also needed something fairly quiet. A 460 seemed to suit my needs best.
It gives a decent speed up to what I had before and makes changes to settings and navigating with higher resolutions 'ok' whereas before it was not really practical.
Here are some numbers for apportioning the total space I have been working out for myself:
4.5m poly[840mb] + A2@300ppi with narrow margins[655mb] + texture/hdri[290mb] + reserved space[260mb] = ~2gb
8m poly[1510mb] + A4@300ppi with narrow margins[155mb] + textures/hdri[100mb] + reserved space[260mb] = ~2gb
I have 8gb of ram and this is fully used loading 8m poly.I think if you had >2gb card you would want 12gb or more ram.
I think the most obvious benefit of 2gb is that it allows a decent number of polys and a large image size together. If you are a texture freak you will enjoy the extra too.
If you need a 3gb Tesla you are a pretty heavy duty user with arch viz needs I think.
6gb is probably for people who want to put in trees and grass in..not sure what the needs are there - not my area of work..
We havent heard anyone on the forums talking about top end requirements.
I would be interested to hear if anyone is finding 3gb too small.
The thing is that atm only 224 out of 336 cores of the 460 are doing anything for the performance you mention. Currently the performance of 2x460 are about the same as a single 480.
Cuda 3.2 due soon is supposed to fix that idle core issue + there is some possibility that some maths functions are working noticeably faster in cuda 3.2 for Fermi cards.
It remains to be seem if/how this will benefit Octane but it could be a decent speed up is coming, perhaps greater than 1.5x if we are lucky so 2x460 would then outpace a single 480.