Originally Posted by l88bastar
*SIGH* Do I really need to go over this?
The 7990 is single card dual GPU crossfire not a dual Card Dual GPU crossfire....they are totally different.
Dual card crossfire eyefinity solutions are terrible because the crossfire bridge cannot handle the information workload needed for the large eyefinity frame buffer. This is why you ALWAYS have crappy microstutter with crossfire eyefinity solutions.
Single card crossfire such as the 7990 does not have this problem as its internal crossfire bridge is much, much beefier versus the flimsly double card options.
Nvidia does not have this problem because they let you run one monitor off of each card and they don't have to worry about overloading the SLI bridge with frame rate reduntancy. This is why Nvidia is superior for multi-card applications.
I dunno if you have some special insight into the 7990, given that it's not out yet, but what you're saying here is basically entirely wrong if applied to every other dual-GPU card that's come out previously.
There's never been a single card/dual GPU situation that was superior to it's equivalent dual card SLI or Crossfire solution when every spec is held exactly equal.
Examples include SLI GTX275 vs GTX295 (at GTX275 clocks), SLI GTX580 vs GTX590 (if you put them at equal clocks), and the last three previous AMD solutions like 6990 vs Crossfire 6970 ... the single card has NEVER been faster than the dual card ... never. It might match it in some cases, but it's never faster, and it's often slower, esp. with nVidia cards.
And all of these dual-GPU cards have functioned identically to regular SLI or Crossfire when it comes the communication taking place between the cores. NV dual gpu cards just use their NF200 chip, for example.
Now, there was talk at one time of AMD having some special uber-inner-connect called Sideport between the two gpu's that they physically put on the 4870x2 but they couldn't get it working and abandoned it as far as I am aware.
Like I say, perhaps you have some insight on the 7990 I don't know about, but in terms of every card that's come before it ... I just wanted to make clear to everyone that what you said there does *not* apply.
Two physical cards have always been faster than, or at worst, equal in performance to, their single card/dual GPU counterparts ... and they've ALWAYS just been EXACTLY SLI or Crossfire implemented on a single card in terms of their inner-gpu communications ... with all the myriad potential 'issues' that SLI or Crossfire inherently have like microstutter and occasional lack of multi-gpu profiles (and I'd bet that if HD3D isn't supported for regular crossfire, it probably won't be for the 7990 either, since that's the point of the discussion)
Also, I cannot see why there'd be any advantage to nV's configuration of running the third monitor off the 2nd card. I don't know what you're on about with the talk of 'frame rate redundancy', but the total 'traffic' along the innerconnect between the GPU's (the PCI-EX bus and the Bridge) should be identical regardless of whether the third monitor is plugged into card 1 or card 2. So I'm curious where you got this information. Sounds like someone's 'speculation' more than it sounds like 'proven fact', let's put it that way
Lastly, microstutter has nothing to do with bandwidth issues created by transferring the frame buffer contents from one card to another for display. Even a very large frame transfer requires very minimal bandwidth (compared to what's available) as it's basically sent like a compressed image file, like a .jpg type of deal. It's nothing compared to the traffic created by the GPU's talking to the CPU, put it that way. AND the buffer is NOT necessarily transferred over the Crossfire bridge, in fact it's much more likely to be transferred via the PCI-Ex bus.
AMD has microstutter issues (in part) because they place "Scaling" at a premium over "Smoothness" ... put another way, increased smoothness has 'cost' in terms of scaling (because it involves one card having to 'wait' for the other from time to time), and AMD has calculated that better scaling improves their bottom line more than smoothness. This is primarily because of the way review sites fixate entirely upon FPS, rather than studying frametimes.
It's also been proven that the closer your GPU's are to max capacity (i.e. 99% usage), the more likely you'll get microstutter, whether it's nV or AMD. So while I don't care to go into detail about why that is so, suffice to say it's not because of a lack of bandwidth with which to transfer the buffer contents ... however, the underlying reason for it means microstutter is more likely to occur at massive resolutions, simply because the cards are more likely to be at 99%.Edited by brettjv - 4/18/12 at 9:24am