Did SLI and Crossfire fail because they had to work with two cards?

[Quartz]

Looking back, it seems to me that the critical flaw with both SLI and Crossfire was that they had to work with two cards when it should have been a minimum of three. With two cards you can get the dreaded micro-stutter, but with three, you can have two (or more) doing the graphical work - alternate frame rendering - feeding to the third which acts as a sync card and outputs a micro-stutter free image to the display. Unfortunately neither Nvidia nor AMD went this route. As a bonus, since the games only interact with the sync card, this would be invisible to the game so profiles would not be needed.

How say you?

 

[Quartz]

But with three cards you still got micro-stutter under the current schema, just less of it. In what I'm proposing, one of the three cards acts purely as a synchronisation and output card.

 

[Quartz]

Seen some interesting information, not sure if it is something that is just an experiment out of curiosity or what, that one approach being looked at currently is some form of having MCM hardware that actually presents as a dozen virtual GPUs with each being divided out bits of the scene with the ability to dynamically allocate the resources each one has on the fly depending on load.

Having each GPU render part of the image was one of the original ways that SLI worked. It quickly fell out of favour IIRC.

 

[Quartz]

Some games still used SFR - you could force it on as well - there were even several variants of it. Problem was actually dividing out the scene somewhat equally for the GPUs.

Ah yes, SFR. I didn't know it was still used.