Chipguy, re: 1) I never predicted Merced's frequency.
Yes my statement was a bit too generic for Merced: You did think Merced would be introduced a heck of a lot earlier than it was. For the rest of the line (Mckinley, Madison, Montecito, and now probably Montvale) you overestimated the frequency. Just to be complete: I do take your estimates seriously as they provide me with some ball park max frequency. Your Opteron perfomance estimate before introduction was pretty good.
Re to my question "So why trust Tukwila?" you wrote: I don't believe I have said anything about the frequency or performance of Tukwila...
I was refering to performance in combination with its release schedule. In response to my question if you thought the lowering of the 2.1GHz introduction frequency for Montvale could be true you said that from Intels perspective it might be wise to spend more resources to get Tukwila EARLIER to the market than to spend those resources on getting Montvale to clock higher at introduction. That shows you trust Tukwilas release schedule to a great extent.
re: Keep in mind that Intel is losing market share in x86 servers and is rapidly gaining share with IPF in the high end
market. Which future products do you think Intel will want to emphasize most strongly in the press?
With regards to server market: Woodcrest/Cloverton and Harpertown. They are clearly emphasizing that right now. The reason must be that they need those products to counter Opteron.
re: [HP discontinued Itanium workstations] because the general purpose workstation is now for all intents and purposes a high end PC with a specialized graphics card.
Why didn't you ask why SGI introduced an IPF workstation?
HP discontinued them because they didn't sell nearly well enough for them. SGI introduced them subsequently because HP cancelled them and they thought that although maybe loss-making it was still an essential enough component to grow their piece of the Itanium market.
re: [Itanium engineers were recently moved over to x86 development because of] platform convergence. The best
people Intel has for this are the ex-ADTers who designed
the EV7 and EV8. They used to be entirely on the IPF side.
Reasonable explanation. I was thinking though that x86 and Itanium engineers have so far exactly the same access to CSI development, but this is actually only a guess. So although I'm not sure about myself I'm not sure yet about your explanation either.
re: [I think all is about right on target and getting better with Itanium] because IPF momentum continues to strongly grow even with Madison generation I2 processors in their final days. All the important OSes/features are on board, ISVs are on board (there are now over 5000 ISV apps for IPF just under HP-UX alone). The new Fujitsu system is very impressive and Fujitsu seems very serious about selling a shipload of them. All this with the introduction of Montecito just a few months away along with top to bottom chipset/system revamps of HP and SGI IPF products to match the new chip. What's not to like?
While all you say is quite true, it is also true that Itanium is currently almost only replacing older architectures whose funding was cut close to nill YEARS ago, and the PA-Risc/Alpha/Itanium/R16000(or whatever the cpu architecture was called that Itanium is replacing at SGI) market isn't growing at all for as far as I can check it (from income statements HP+SGI, and recent market share reports). The tier 1s that you mention to be on board almost all also shipping x86 and other architectures. Power is not about to be endangered by Montecito. SUN has a strong following because of their software and their recently more or less acceptable roadmap because of their collaboration with Fujitsu and AMD, and their customers are not expected to jump across to Itanium. Anyways, from my point of view Itanium hasn't been strong enough to grow the PA-Risc/Alpha/Itanium/R16000 market so far, and is not likely to significantly enlarge it in the near future, but as always I can be wrong.
You only talk about platform convergence. Does this mean you don't expect the cpu architectures will converge within the next four years? And if so which architecture do you think the convergence will be based on, x86/EPIC?
Regards,
Rink
