Quote Originally Posted by zerazax View Post
A lot of those things are driver issues and people playing the game where the card is obviously choking. Here's the main thing though:

Multi-GPU is a very recent thing, despite being an ancient idea. Sure, the Voodoo cards tried it out and after Nvidia bought out 3dfx and adopted their own version of SLI, it took a long time for it to mature to where it is.

Look back one year and see where we've gone since then. We've finally had multi-monitor support (CF), Hybrid SLI and CF for both power saving and for boosting integrated GPUs, and improved scaling (as the 9800GX2 shows that it is still quite formidable even against the new architecture with a single GPU).

As both sides get more committed (remember, Nvidia really started the experiment with the 7950GX2 and then continued it on with the 9800GX2) to promoting and selling their multi-GPU setups (why else would Nvidia want to hold so much over SLI being exclusive to their chipset division), they've finally seriously added features and driver improvements that we could only have dreamed of 2 years ago. Heck, even a year ago.

And anyways, I feel that you are envisioning multi-GPU's too much as simple on-board SLI or CF. I think both sides are exploring different ways to make GPU's communicate. For example, it isn't out of the reach to say that Nvidia may one day adopt an interconnect similar to a memory controller in order to improve the capabilities of their cards for GPGPU purposes and really hold to their claim that CPU's are obsolete.

Obviously a lot of that is speculation but seeing as to how both Nvidia and AMD are pushing towards integrating GPU's with CPU's (in other words, combining the two to make on uber processor), they will certainly have to come up with new technology for these things to happen.
I understand your point, but a lot of the problems with multi-gpu are just "the nature of the beast". You'd need to drastically change the fundamentals of it to improve things... it would need to be many chips able to function as 1 big chip, which is pretty damned hard to do.

Until you've programmed a game to be scale well for many-gpu, you can't sit here and tell me that it is the future. It's a hassle and a hack that is a necessary evil to be #1 (if you don't do it the competition will), just like the cold war's nuclear standoff. The point of future technologies are to simplify things, not complicate them. No game developer wants to add worrying about multi-gpu scaling to their list of " to do today."

The future is to not have games be required to code things a certain way to get the scaling.. it should work perfectly every time, which means that AFR goes out the window.