[citation][nom]A Bad Day[/nom]Altering an architecture to fit in more stream processors, texture units, and etc requires millions of dollars for drawing up the plan, testing, and then modifying the fab machines to build the new dies.It's cheaper to build a big die and bin it accordingly, because manufacturing is never perfect. You're always going to have flaws ranging from one that requires a little extra voltage to one that only has one functional stream processor.And the second part of your answer is essentially CrossFire, that's a 7990.[/citation]
They don't need to alter any architecture. The architecture itself remains unchanged. Besides, like I said, AMD could simply take two current dies and add a very wide and high frequency Hyper-Transport connection between them. Why would AMD need to make any changes to the machines for this, yet not need to make changes for making the big die? That doesn't make sense. The second part of mine is not essentially CF, it's different. CF is two separate chips and my what I said is making a dual die chip. It's not the same. It's more power efficient and better-scaling. The 7990 has two separate GPUs, not a single GPU with two dies.
Also, as dies linearly increase in size, the problems associated with binning increase exponentially. This is why AMD dislikes big dies and probably related to why Nvidia didn't want another compute oriented architecture that needs huge dies to be competitive in high-end gaming performance when they made Kepler based video cards.