Tony, Re: "Sun has, what, UltraSparc IV and V in development, so they crank those out and then stop any more Sparc development? I looked at their website and all the microprocessor design job descriptions mentioned Sparc, didn't see any hint of Itanium at all. Well kept secret what you say about the Itanium platform work? Of course, that investigation would be more architecture than design. A guy I know that's made the rounds of a few computer companies in the valley, the other day used the words "Sparc is dead". We'll see."
Here's the thing. UltraSparc development has been really slow. I remember seeing slides that mentioned USIV in 2001 and USV in 2003. USV was supposed to be multi-threaded and multicore, but now there hasn't been any mention of it. Equally alarming was that Sun doesn't even seem to have any signs of USIV in the works.
Many expected to see information on the USIV at this year's Microprocessor Forum. I was there, and all Sun had to describe was information on the USIIIi. Now, keep in mind that USIII was originally supposed to be a chip that launched in 1999. It was delayed until 2000, but was supposed to launch at speeds of 600MHz, 750MHz, and 900MHz. Unfortunately, due to bad yields or bin splits or whatever, 900MHz USIII chips were unseen until midway through this year. Now, Sun is using a smaller .15u process, and all they can seem to manage is 1.05GHz USIII chips, and barely any systems using them.
The 900MHz USIII is only now being implemented in Sun's large server systems. The USIIIi, now set for next year, will supposedly be able to clock up to 1.4GHz (and even that is doubtful, IMO), and only usable in up to 4-way configurations. That means that Sun's 900MHz, and later 1.05GHz USIII chips are all that they will have available throughout next year for their large processor systems. Compared to 1GHz McKinley chips that are likely to be far more powerful per clock, Sun's chips will only seem that much more underpowered.
Going into the future, Intel's Itanium roadmaps are very well laid out. Madison is set to launch in 2003, as well as the lower cost Deerfield. Sun will have a tougher time keeping up, and even if they happen to get USIV out in 2003, how much will it gain them, and how much will Sun have to continue investing to ensure that they will have competitive CPUs?
I think that going forward, Sun will eventually see what it was that convinced Compaq and HP to go with all Intel. Mostly that epiphany includes the realization that investing in their own architecture does not guarantee as much of an ROI as buying chips already designed and validated by another company. Compaq and HP have declared that most of the differentiating factor in a large server system comes not from the CPU, but from the platform, which includes the chipset and other components.
Compaq and HP will continue with their platform development, and so will Sun, but the question is whether it's still worth it for Sun to push their own CPU. After all, if they can do the same things on an Itanium CPU - if they can port Solaris and build their highly prestigious platforms towards Intel Architecture - then what's the difference? It might be less expensive for them to do that in the long run.
wbmw |
