Advanced Micro Devices Inc (AMD)

[dougSF30]

Yamhill demo at IDF. Might be in Nocona, maybe not until Tejas.

BTW, this woman is a tech reporter who goes waaaaay back in the valley. She KNOWS lots of things.

http://www.linuxworld.com/story/41467.htm

---------------------------------------------------------


Intel to Out-Shy Yamhill
January 29, 2004

By Maureen O'Gara
Page 1 of 1

Since Prescott, the next Intel desktop chip, is coming out on Tuesday, and since Yamhill has been described as Prescott on 64-bit steroids, we asked Intel about its alleged skunkworks answer to the currently styled AMD64 hybrid 32/64-bit architecture.

Instead of weaseling around about whether Yamhill exists or not, as Intel always has, Bill Kircos, an official, high-ranking Intel spokesman, conscious of the subtle change in tune, said that Intel would bring out an x86 chip with 64-bit extensions like AMD's if "customers demand it and the ecosystem for it develops."

Intel president Paul Otellini said much the same thing in a Webcast on Wednesday.

That ecosystem would necessarily be an AMD-established ecosystem and Yamhill would have to be able to run the same operating systems as AMD64. So the market wouldn't bifurcate.

Linux runs on Opteron, but the AMD64 and the presumptive Yamhill are more a Microsoft play and Microsoft reportedly told Intel months ago that it's not going to build two 64-bit x86 operating systems.

According to the latest schedule Microsoft's OS for Opteron, once due last year, won't be out now until the second half in the first service pack for Windows Server 2003.

It's also been common gossip around Intel for ages that, if the company were to enter the market, it would start with a server Yamhill, the requirement of 64-bits on the desktop still being a ways off.

The server version of Prescott is a chip code named Nocona due shortly after Prescott, say, Q2.

Now, Intel watchers like Insight 64 principal Nathan Brookwood and Merrill Lynch analyst Joe Osha believe that the 64-bit extensions in Prescott and Nocona are incompatible with Opteron because when Intel locked down Prescott and Nocona AMD hadn't published anything about its 64-bit extensions.

[ Very interesting hypothesis for those "dark transistors", no? - Doug ]


The desktop chip after Nocona is Tejas, and, by their theory, Intel's first opportunity to bring its 64-bit extensions in line with AMD's. (Wow, now that would be a turnabout for the record books.)

Tejas is reportedly due late this year, early next year, a very quick crossover from Prescott, possibly a noteworthy point, but for Intel to turn on the 64-bit extensions in Tejas it would theoretically need a commercial 64-bit operating system and Microsoft has yet to even promise an AMD64 desktop OS whereas Linux on the desktop is still hype.

That suggests that Nocona is still the likely horse to watch since Intel has lots of friends at Microsoft who could tweak the upcoming AMD64 operating system to accommodate any architectural differences. Anyway, how hard is it to bolt 64-bits on to a processor? Adding it is a fairly straightforward exercise and there aren't too many ways to do it regardless of operating system support.

Which brings us to the next Intel Developer Forum in mid-February where Intel is going to give a Yamhill technology demonstration, although it could be jeopardizing the 64-bit Itanium chip that cost a trillion dollars to get to market, and is only selling modestly.

The event will make AMD's day and validate its hybrid approach.

Intel would neither confirm nor deny reports of the demo, but informed sources say it's true and of course the timing is exquisite. Long about then, Intel's hereditary enemy Sun Microsystems should be announcing its promised Opteron line - which it's reportedly already delivering - and Itanium co-developer and primary booster Hewlett-Packard should be confounding people by adopting Opteron too.

That just leaves Dell for Opteron to scoop up.

IBM already has Opteron boxes, but they're a limited HPC experiment because what IBM really, really wants are Yamhill boxes that don't disrupt its glorious Xeon sales - it's currently the biggest Xeon seller - and force its Xeon base to forklift on to the Itanium and Itanium's x86-incompatible EPIC architecture simply to go to 64-bits. If IBM wanted to do this, it's said, it would forklift them on to its own 64-bit x86-incompatible Power chip.

(IBM currently wants to push Power chip into the Linux/HPC market big time and it's probably fair to say that IBM has little use for Itanium because it threatens its core Power business.)

Naturally Intel will have to position Yamhill technology for low-end servers and Itanium for the high end and get people to believe that it can do two seemingly incompatible things at the same time without wrecking the delicately poised Itanium.

It would have to identify Yamhill as the future of the great and mighty Xeon line that has captured 85% of the server market but only 45% of the server revenues, and cast Itanium as the vehicle to capture the other 55% of the revenues that are still going to RISC systems.

Illuminata analyst Gordon Haff speculates that the only reason that HP would adopt Opteron is that it was forced to by strong customer demand and he equates that demand not with Opteron's 64-bitness but with the fact that the widget - thanks to its added registers, integrated memory controller and HyperTransport links - boosts 32-bit Windows apps.

Haff says, "To be sure, Intel's CPUs have their own performance technologies, such as Hyperthreading, but it's by no means a given that Yamhill technology will offer the same sort of performance boost relative to 32-bit Intel processors that Opteron does relative to AMD's own 32-bit Athlon CPUs."

Which leads him to think that "Yamhill may not threaten Opteron much in the near-term unless it combines very strong 32-bit performance and price/performance as well as extensions to 64-bitness."

An interesting point, but one gathers Intel is pretty confident.