POST OF THE DAY
Advanced Micro Devices
100 MHz Itanium competing with Hammer

Format for Printing

Format for printing

Request Reprints

Reuse/Reprint

By ValueNut
January 24, 2001

Posts selected for this feature rarely stand alone. They are usually a part of an ongoing thread, and are out of context when presented here. The material should be read in that light. How are these posts selected? Click here to find out and nominate a post yourself!

I think I am not the first to post a reference to this, but I didn't notice anyone responding to the article in the way I expected.

Here is a Register article that documents that the 667 MHz Intel Itanium will run 32-bit code only at the speed of a 100 MHz Pentium.
 
They conclude by saying that "in an irony our friends at Transmeta will surely savour - Crusoe's software emulation of Intel's 32bit instruction set is faster than Intel's hardware emulation. In fact, Crusoe could even emulate the IA-64 instruction set, ".... while using less than 5% of the electrical power. :-)"


What's going on here? Why would Intel embarrass themselves this way? Didn't they see it coming?

I can't imagine how anyone in marketing or R&D could allow 32-bit emulation this slow to be allowed into the design. As the Register implies, they could have left out 32-bit emulation completely and been better off. They could have saved the die size that was required to do the 32-bit emulation and paid someone to develop a 32-bit emulator in software that they could have given away free with the chip.

It's clear that what they did made neither marketing nor engineering sense.

Imagine what must have happened here when first silicon was being tested:

Validation Engineer: Hey, Mr. Marketing, I've got this test report on the 32-bit emulation for the 64-bit Itanium. I just wanted to --

Marketing Director:
Oh great! Did it pass? Are we ready to ship yet? We're gonna be the laughing stock of the tech industry if we slip one more time on anything.

Validation Engineer:
I just wanted to say that even though you told us it doesn't really matter how fast the 32-bit emulation is, I thought I should point out that it's kind of slower than we expected, and I think we should do something about it -- like optimizing the microcode or maybe turn off the emulation. We could even make the chip smaller by leaving out --

Marketing Director:
Why? The spec here says it's gonna be software backward compatible with IA32. It's gotta be or we might not be able to move the product down market and sell it to people who think they might want to run 64-bit later and pay a higher price for that privilege now. I'm always having to remind you engineers how important it is that we can keep our margins up.

Validation Engineer:
Yeah, but --

Marketing Director
: Look, even if it's a bit slower that'll just encourage people to move their applications to 64 bits even faster. Right? I mean, really, what other choice do they have? After we kill off the IA32, they'll have to buy our chip and they'll have to upgrade all their software to 64 bits. Then we've really got their family jewels in a vice because we've patented the IA64 instruction set! The plan is brilliant. We'll be a monopoly again, sooner than you can say "Intel rules!".

Validation Engineer:
Yeah, but if you knew just how sl --

Marketing Director:
Hey, look. I've got another very important meeting right now. Somebody will be waiting outside my door to talk to me any second now. Can you just red mark the important bits and leave your report on that pile of other reports. I'll read it just after my meeting. Promise! Bye!
Is that the way it happened?

I think this is a bigger advantage for AMD than many realize. If AMD comes out with the Hammer on schedule and performance is better than the Athlon, then AMD will dominate the market for power users who want to keep running a 32-bit OS, but want to be able to upgrade to 64 bits whenever it makes sense. Until I saw this article, I did not realize the extent to which Intel had abdicated that opportunity. The power-user market is a very nice market. Though it may be small, it has excellent margins.

Obviously, Intel went through a period of time where their strategy involved killing off the IA32 by replacing it from the high end down with the IA64. Since the IA64 is effectively protected from cloning this would have given them a new monopoly.

Their strategy required an IA64 with higher performance than their existing IA32 family and no real competition in the IA32 market. Neither one of these things happened the way they expected. Any decisions they subsequently made based on these premises have gone wildly wrong.

When did Intel realize they had made a big mistake with their strategy? Was it last summer when they announced that basing their high-end entirely on Rambus memory had been a disaster?

How quickly can Intel recover with really excellent designs? The Pentium 4 is a disappointment and I believe the entire IA64 family inherently requires so much memory bandwidth it will never be competitive with their own or any other IA32 products for users who don't need 64 bits. The Hammer doesn't have this memory bandwidth problem because it does not have a code density that much different than older x86 designs.

In short, I am asking, how long will it take Intel to produce a completely new 64-bit design to counter AMD's Hammer product if they started on it as early as one year ago? What is Intel's design cycle time?

If Intel can't do this within the next year, is their any hope they can win a price-performance battle with Hammer? Both Intel and AMD will be at 130 nm by then. I am assuming Hammer's schedule does not slip significantly. That is certainly possible, but I think it's not all that likely. Hammer is really just a souped up version of the Athlon with wider registers and a few new instructions.

If I were Intel, I'm not sure what my best strategy would be right now. I want to get a good idea of what that best strategy is because I am trying to come up with a long-term model of competition in the x86 market.

Since many of AMD's advantages are short-term I want to know whether I should sell AMD on the next runup in price or hold them for the longer term. If Intel is likely to be able to take back all of AMD's recent gains on the long run, then I don't want to hold AMD for long. It looks as if Hammer has a sustainable advantage against the Itanium and the Pentium 4. The Pentium III can probably be tweaked to compete with the Duron. It looks to me like AMD needs to retain the high-end lead in order to preserve a 20%-30% market share.