Integration Fever . . .

Intel plans to merge CPUs and GPUs within a few years. AMD plans to merge CPUs and GPUs in a few years.

Even nVidia wants to get into the act, but they’re different. They want to merge GPUs and CPUs within a few years.

This is all very nice, but nobody has seemed to ask or answer a very simple question:

What’s in it for us? What good would this do us any time soon?

All these gung-ho folks have been curiously quiet about this. We’re not even getting BS about it.

Now I can easily see why a CGPU would be a good idea for those developing it. It would mean simpler motherboards for cheaper desktops and notebooks, and eventually would be a necessity for a phoneputer.

But what about everyone else? You just have to assume that there’s going to be a huge gap in performance between one or two tiny graphics cores sharing memory with the CPU and tomorrow’s fire-breathing dedicated video dragons. It may well be good enough for the integrated video crowd, and that is a very big crowd indeed, but what about the rest?

What if you don’t want a CGPU? Let’s say you want four CPU cores a few years from now. Not three, not two, four. Will that be possible, or will you have to take fries with your CPU order even if you’ll never touch them?

Is the rush to a CGPU to at least some extent a desire to give some of those extra cores something to do?

Folks can talk about Torrenza and other integration products, and so long as they mean merely extra-fast interconnects, that’s just building a faster mousetrap.

But if you carry all this integration to its logical conclusion, what you see is the boutique CPU, a wide variety of CPUs made to order.

That may sound great, but today’s CPU fabrication is built from the ground up on mass production and economy of scale, and it’s more than a little more expensive to come up with 27 flavors of CPU than ice cream.

And how much more will those chips cost? OK, there will always be plain vanilla and chocolate chips, and there likely will be niche specialist chips for high-value computing that will be bought by those happy to pay an arm and a leg for them.

But what about the middle? For instance, will we have gamer chips, and how uhhh, profitable are they likely to be?

Yes, it’s very early in the ballgame, but one gets the impression lately that there’s a lot of specific solutions in search of general problems, and that the real goal of a lot of this is to maintain or justify higher prices at a time when the marginal cost of a CPU would otherwise drop into the nominal range.

Ed