Cores, accelerators, and processors—oh my

Posted: 09.29.08

It’s a quality issue…

We now have four offerings for the consumer space of multi-core chips, which are euphemistically called processors. That is just a taste of how confusing this is all going to be, especially when the market spin is added. For example, AMD’s four core chip called Barcelona is called A processor, as is Intel’s i7 Nehalem. But those “Cores” are in fact full-function multi-stage out-of-order 64-bit double-precision floating-point processors. So the term “processor” has come to mean chip, and the tem core has come to mean processor…a rose by any other name. It doesn’t really matter what we call these things as long as we all call them the same thing and share a common understanding of their definition.

Sticking close to common usage and not inventing or adding to the litany of words for the lexicon will be an underlying goal, if not rule, for this discussion.

Then, we have GPUs with hundreds of 32-bit double-precision floatingpoint processors, but they’re not called cores. They’re called streaming processors and sometimes SIMD arrays. And the competitors of GPU suppliers call the GPUs, application-specific integrated circuits (ASICs) or “accelerators.”

The GPU suppliers openly admit that within their GPUs there are indeed some accelerators, most notably video-processing accelerators for MPEG. There are also texture processors.

Intel, as you know, is coming out with a multi-core, or a many-core processor called Larrabee, which will have between 16 and 64 cores (depending upon SKU). Since the cores in Larrabee are x86 units with a floating-point processor, they are called “cores,” by this new unwritten law we’re developing. However, Larrabee also has hardware accelerators in it for texture processing, so does that make it an ASIC or an accelerator? Not if Intel’s marketing folks have anything to say about it—and they will.

And so we enter the newest battle of words, the newest battle for mindshare and, of course, the all-important battle for market share.

I made a comment a couple of months ago that it was ridiculous to compare Larrabee to a GPU. I was quickly challenged on that remark by an Intel person and asked how I could make such an uninformed statement. In public, no less? My response was how can you compare a 240 processor Nvidia GPU or an 800 processor ATI GPU to a 16- to 64-core Larrabee. There’s no comparison—the GPUs have between 40 to 60 times as many processors. My challenger smirked, somewhat snidely, and said, “Well, I guess it depends on the quality of the core doesn’t it?” I pondered that but never resolved it.

Later, I was speaking to an Nvidia person and said ATI’s new GPU will have better performance. The Nvidiette quickly snapped and said, “What do you base that on?” Well, I said, for one thing it has 800 processors and you only have 240. “Well,” said the Nvidiette, “I guess it depends on the quality of the core doesn’t it?” I thought, there’s an echo in my head, I should see a doctor.

So, now not only do we not have a common term for a processor, or a core, or an accelerator, but along with it, we seem to have a quality issue to contend with.

Of course these are just the opening barrages of the three giants, trying to confuse and affiliate the masses to their cause. That confusion, or spin, will have some effect, because all of the consumers will be the blind men following the one-eyed chip makers. And a bunch of them will buy stuff from all three. And in time, perhaps it will take a year, the logical and correct winner will emerge; the company that is providing the best performance for the best price with the least heat and space.

All three will tell you it’s them, and it’s now, or very soon, trust me.

So, we will have to swim in these choppy waters looking for a life raft and not knowing what to call it when we find it.