More Cores?

The Goldman Sachs report we mentioned earlier today also mentions that AMD “potentially” may make a quad-core processor in 2007.

Let’s hope not.

This isn’t just an AMD thing, what I have to say is just as true for Intel or anyone else doing the same.

Why do I say that?

I say that because multi-cores are an admission of defeat by the CPU companies. Multicores for mainstream CPUs is like suicide bombings by terrorist. You don’t do it unless you have no better option.

Yes, multicores have their legitimate place in the computing world. A small one. The desktop really isn’t that kind of place.

Yes, multicores can sometimes give you a “smoother” ride than a single processor, but it’s not something the average person isn’t going to pay much if anything more for.

Multicores are inherently less efficient for most desktop activities than a faster single core. First, many computing tasks just can’t be split up among many hands.

Second, even if you can split up a single task, then you have to have coordination, and that chews up more and more cycles as more and more players join in. I suppose you could have sixteen processors split up your gaming screen into tiny pieces, but having sixteen processors all trying to access GPUs and memory spaces simultaneously sounds like a traffic jam waiting to happen. That probably can be avoided with enough spotlights and traffic cops, but wouldn’t a single bus be more efficient than sixteen cars with single drivers?

If You Can’t Make Better, Make More

The craze for multicores has been caused by the CPU companies hitting a wall when it comes to making their products faster (at least within the PC environment).

Only after hitting that wall did “two heads are better than one” become a cry.

It’s not a better approach; it’s a better-than-nothing approach.

It reminds me of a much older technological competition: the U.S.-Soviet space race.

After World War II, both the U.S. and the Soviets essentially futzed around with Nazi V-2 technology, not doing very much with it.

Then one Soviet rocket scientists essentially said, “You know, if we strapped a bunch of these V-2ish rocketa together, we could actually get into space with that.

It was a quick-and-dirty and cheap way to get into space ahead of the technologically advance but lackadaisical Americans, so they did it, and we got Sputnik.

If strapping a few together was good, strapping more together would be even better, so the next thing you knew, the Soviets got the first man into space, too.

The Americans were 0 for 2, so the U.S. President at the time decided to try to hit a grand slam by promising to put a man on the moon by the end of the sixties.

The Americans and Soviets took rather different approaches to this problem. The Soviets kept with the tried-and-true approach of using many low-thrust engines. The Americans (well, headed by one ex-V2 German guy) went back to the drawing board and spent five years designing one big rocket with some big engines.

You know who got to the moon. This was a big reason why.

So when AMD (and Intel, and for that matter nVidia or ATI down the road) talk about multiples or multiples on top of multiples), this is an admission of technological failure, not success.

If a breakthrough is found that puts an affordable hole through the current technological wall, this talk of multi-multis will fade, and it will fade because it’s not an advance.

It’s a kludge.

Ed