Tejas To Be A Modular Processor?
Here’s the article mentioning this.
The concept itself is fascinating. In theory, you could “build your own CPU,” including the things you do want, and exclude the things that you don’t.
If you’re Joe Sixpack, or his mom, you can go with the minimum. More rabid folks can get (and pay) more.
Please note what this scenario would do eventually, though. It would free Joe Sixpack from the geeks.
As it stands now, the geeks have essentially dragged Joe Sixpacks along the road of faster CPUs, and having Joe foot most of the bill. The advanced R&D got spread among all the CPUs, whether used or not.
What modular CPUs do is not only isolate basic and advanced CPU functions, it also isolates their cost. If, for instance, Intel makes a core 32-bit processor, gives people a 64-bit option, and only 5-10% want it, those who want it are likely to bear most or all of the costs for that technology. Not Joe.
That would mean cheaper processors for most people, and significantly more expensive ones for those who want or need more.
That’s good for Joe. It’s no big deal for the professionals for whom time is money.
For those inbetween (like most in this audience), you’re caught between a rock and a hard place.
We’ve thought for years that the PC market would eventually split into two much separate markets: the PC as a cheap consumer electronics good for most, business-as-usual (but with much higher prices) for those who need more than that.
This is another step in that direction.
However, Intel’s near-term intentions are probably rather more modest.
How To Do A Multicore
I suspect that this is going to be Intel’s way to implement multicores. What should be noted is that multicores can be implemented a number of ways, and do different things.
First, there is the simple dual-processor concept. You have two identical chips working together on the notion that two heads are better than one.
I don’t doubt they’ll do that. Indeed, we may very well see something like this within a couple years: Intel puts together two cores running at relatively low-speeds, and then claims to have the equivalent of a 8 or 10GHz processor.
They’ve made comments hinting at that in the past, and this modular structure could well be how they’ve planned to it.
That’s not the only possibility, though.
The article linked above mentioned the possbility of a 64-bit add-on. As the article put it, “The idea seems to be that people can buy a 32-bit module, and then add in the 64-bit processor.”
Hmmmm. There’s no need to have a separate CPU if you’re going to have x86-64, all you need is a few million transistors. Just stick them in the CPU and be done with it if that’s all you’re out to do.
Nor would there need to be a different version of Windows; the only change MS would have to make is to change the name of Windows for AMD64.
But what if that 64-bit add-on is an IA64 processor? Then all of this makes a great deal of sense, at least in Santa Clara.
Such a chip could knock out many of Opterons’ advantages and Itanium disadvantages. Such a chip could outperform Opterons in both 32- and 64-bit operations, and would let people move from 32-bit to 64-bit operations at their own pace with an add-on.
Elements for Windows would need to be an OS that could seemlessly handle two separate processors. a little different task than the 64-bit Windows we have today.
And how much would such a chip cost? Well, that would likely be a big problem, now wouldn’t it? 🙂 Certainly more than the average desktop user would pay.
Even if this were merely a CPU that could accept a second processor core, 64-bit or not, this would be one of the if not the most revolutionary chips Intel has ever come out with.
However, Tejas isn’t supposed to be a “revolutionary” CPU (that meaning a redesigned from-the-ground up processor core). The next CPU after that: Nehalem, is supposed to be that.
It would make a lot more sense for Nehalem to be that revolutionary chip.
Perhaps Intel will split the revolution a bit. They might start with some server versions of Tejas chips, then if things go well, make this a standard feature with Nehalem.
Since Nehalem is supposed to be the last Intel x86 design, it would make sense to allow people to start transitioning between it and IA64.
Inside Intel’s Head
The idea is to figure out what Intel will likely do. The best way to do that is to see what they’ve done in the past.
Intel is stubborn to a point shy of suicidal. If they get fixed on a notion, like RDRAM, they don’t change their minds too easily.
Intel has had far more reason to get fixed on IA64 than they ever had with RDRAM. Do you think that when Itanium is finally getting its act together and showing signs of delivering on its initial promise, Intel’s going to drop it because AMD sold a few Opterons?
That doesn’t preclude using x86-64 as a stopgap for Xeons if Opterons do too well, no more than Intel hesitated to spiffy up a Xeon and toss that into the desktop fray.
Would that have to be a big win for AMD and loss for Intel? Not at all, and here’s how.
People have assumed that it has to be either x86-64 or IA64 for Intel. That’s not true at all. It can be both.
Intel has the inherent right to x86-64; x86 is their property, and AMD only can use that due to a deal made a long time back.
AMD, on the other hand, has no current right to IA-64; it only has IP rights to x86.
There’s nothing to preclude Intel providing three ways to skin the cat. If you’re doing a IA64/IA32 hybrid, it would be a trifling addition to make it into a IA64/IA32/x86-64 total package and make it swing all ways.
Would MS support such a thing? Sure, why wouldn’t they? They have Windows for IA64 already. They’re building Windows for AMD64. All they’d have to do is get the two to play well together, which would be only a little harder than getting 32-bit XP and IA64 to do so.
It would be a lot easier for Intel to provide a “universal” chip that just happens to support x86-64 rather than swallow x86-64 by itself.
Nor would Intel have to provide IA64 immediately with a modular chip, either. They could provide x86-64 today and provide IA-64 later.
If Intel did that, all the seemingly contradictory stories floating around stop being contradictory. You could buy an Intel chip with x86-64 to begin with, then upgrade to IA-64 if you chose later.
Something AMD can’t offer.
I’m not saying Intel is going to do this, but if the statements in the article are true, they certainly make it possible for Intel to do this.