FlyerDragon

Wider memory bus means just about nothing. It's all about memory bandwidth. Titan Black has THE largest stock memory bus on the market, at 336GB/s, despite the 384-bit mem bus. The R9 290X only has 320GB/s memory bus, due to a lower memory clock even with its 512-bit mem bus. A lower graphics card being unable to push a game's high settings (which happen to use a lot of vRAM) is just the card not being strong. But you could use a lot of memory without actually hitting the card heavily. A good example would be Titanfall; which uses 3GB+ of vRAM at 1080p once you set texture resolution to Insane. Even if the rest of the game is on minimum graphics.
 
Anyway, it DOES take more memory bandwidth to fill a larger buffer in the end; it's true. Obviously filling and emptying a 6GB buffer is more demanding than a 3GB buffer etc. BUT if your application is one where more vRAM = better no matter what, you're probably going to benefit more from a memory pool that is slower to completely fill but larger overall. Like I said though, you should check up on the programs you want to use to make SURE that a large memory buffer is what you're after. OR if you really want to be able to use CUDA. If you are aiming for DirectCompute/OpenCL calculations in your programs, AMD is your choice. They are FAR better, and their popularity for bitcoin mining is a direct tell to that. Even the CUDA miners don't help nVidia cards catch up to AMD ones for that kind of thing. So yeah, do your research, then pick your card.
 
Anyway... as far as GAMING goes, you can easily use up 4GB or so of vRAM depending on your game and setup and OS. If you want more detail about how vRAM affects gaming, you can read THIS and then make your decision. As for pure gaming capabilities, the R9 290 and the 780 are going to trade blows. One is better in some games, the other is better in others. Pick which card vendor you like.