Kapkan, Re: "Unlike scientific apps that can't be easily broken into separate tasks, most server apps don't really care if you use one I2 or 2.35 PIIIs."
More machines means more maintenance costs, and not every app scales well over distributed loads. There is also more administration costs, more software licensing (can be *very* expensive), more complicated installations, more points for failure, etc. Sometimes, businesses prefer to consolidate their systems into a few powerful servers, rather than racks full of commodity chips. Then again, some businesses prefer the racks, so there seems to be a market for both solutions.
Re: "Rack space and power consumption are much more important."
Yes, but not so much as you think. The so-called "low power" server market that Transmeta was trumpeting a year ago seems to have lost momentum. Sure, people want lower power, but they seem to want performance, more. But again, it's useless to generalize. I think there is a market for both.
Re: "You can fit 18 PIII blades in a single 3u 19 inch rackmount. I don't think that I2 can match the computing density of x86 for two more generations, even if x86 was standing still."
IBM is planning on putting Itanium 2 in a "blade" server early next year. Beats me how they plan to cool the thing, but Power4 will even be joining that market as well. You might be surprised the solutions that vendors *already have* that can accommodate devices with very high power dissipations. Also, one of the 4 servers that HP just released for Itanium 2 was a 2U rack-mountable dual processor server. 2U and 3U racks tend to be the "sweet spot" for mountable servers.
Re: "There is not a single compelling reason for anybody but Intel to push I2 into the server space."
I think there are many compelling reasons, but fewer for people already invested in x86 based servers. There seems to be a lot of cost involved in abandoning that infrastructure. On the other hand, remember that 60% of server revenues out there are from non-Intel based solutions. That's where Itanium 2 is being aimed, and in that market, proprietary solutions already have a small, custom software base that is already in progress for being ported to IA-64. Compared to those proprietary solutions, Itanium 2 is much higher performing, and much lower cost. I'm not making this up - just check the benchmarks.
Re: "It is obvious to me that IA-64 will fail."
You list some very valid concerns, but I don't think any of them create a null set of potential customers. However, I do agree that any progress with Itanium 2 is going to be slow, especially in the current depressed market. If anything Itanium 2 will fail because of market conditions, but at least infrastructure will be ready for Madison. I don't think there is anything fundamentally uncompelling with the architecture. I think it is very much competitive with Power4, which has already gained a lot of support with many of the same problems as Itanium 2, except that the Power series does not have as extensive of a future roadmap, IMO. Intel will probably continue to push the Itanium series for as long as they believe in it, and if their roadmap is any indication, they really do believe in it.
But maybe you're right, and Intel is too good at being a volume silicon provider, and not so good at catering to a high end niche. If that's the case, then all this Itanium development is just a waste of spending. However, I see no evidence to suggest that, yet. Merced failed because it was a dog, and it had virtually zero software support. McKinley performs great, but software is still limited, systems except for HP's seem to have been delayed, and the market isn't spending as much money as it used to. So maybe McKinley has in store the same fate. But if you are going to blame it on the architecture, I think you'd be selling it short (no pun intended, given that it's what you do best <g>). So even if McKinley does fail, I still think that IA-64 in the long term will be a successful market for Intel to get into.
wbmw |
