k1114 :
I had to leave before I answered the other stuff in the last post but here goes. I don't see how any aa method will help with this issue. But there is no extra type of aa with multiple gpus.
You can't "hyperthread." I don't see how this is relevant either.
Fillrate = core clock x texture unit or rop. It's useless info imo.
I think you are overthinking the issue and complicated it with bus width, extra info and trying to give odd examples. You simply wanted to know if 2x 270x 4gb would be better than 2x 270x 2gb and if it would stack up against a 290 4gb. Ok so you can't afford a 290 right off the bat, then get a 270x with the vram you need for cf later. Playing at 1080p, 2gb is fine, the person might have just wanted to run 144fps so the extra vram is useless. In this case, the 290 just has extra vram for nothing. Only if you are higher res will getting the extra vram be useful. The single 270x 4gb will obviously not have enough power to run that by itself but you obviously are buying 4gb to cf later for that purpose; to be similar to a 290 4gb.
thank you for your input, sorry for the convolutedness of some of my questions and examples... thus thanking you for your patience as well.
as for multi gpu aa; http://www.nvidia.com/object/slizone_sliAA_howto1.html, and http://www.nvidia.com/object/quadro_sli_fsaa.html, knew i had heard somthing like that. dosent looks like it applies but knew i had read it, and any aa method adds to that frame buffer, thus vram usage, but is useless if the card cant push it, which is why most reviews limit themselves to 4x aa, at that point you are trading framerate for image quality by a good margin (depending on setup), and as you said one 270x can't push 4gb, but can 2?
yes, i am aware of no 'hyperthreading' on gpu's, thus the quotes and it was just an example referencing intel's method of splitting the core in 2 for more threads, with 5800 something cores in a 290x that would be one heck of a hack job...... odd example(back of the hand slap, ouch...)
so far you are understanding the question the best; does the extra vram help in xfire?..... would it matter or are these companies just wasting the memory chips just to earn a few extra bucks? as an example, toms hardware review of 2x r9295x2's in quadfire, it repeatedly mentions hitting the vram ceiling on a few games, so in that case, hypothetically speaking....., what would it need? that's sort of where i thought of this. i think without any real world tests (hint hint tom's) which i agree are the best, this is just a bunch of speculation and thats that.
why there are no benchmarks is a good question but i like 'pushing the envelope' of whats possible and this site has answered many questions over the last couple of years ive come to it, much respect for its methodology and mindframe, and enough so that i finally joined when i thought of this just because SOMEONE on here might have an answer thats definitive, have that type of setup and be willing to run some gpuz tests. at this point im not going to pick a best answer, but i think id like to leave it open just as a sort of sticky until a benchmark is made, but between my bad diction and your noteworthy answers what else can be said until said benchmark. thank you for your time k1114 and when i win the lottery or that rich uncle of mine that i dont know about leaves me his inheritance i will waste the money JUST to answer this and let you know, real world style