Chipguy, Re: Market data I heard to date shows that the designated RISC replacement market (PA/Alpha/MIPS) represents less than half of IPF sales. IPF is taking about the same chunk of opposing RISC business as friendly RISC business with the remainder coming from x86.
I don't know the quality of your resources and still am slightly doubful half of IPF sales is from opposing RISC business. Overall picture I have in mind is this:
HP quarterly Enterprise and Storage revenue barely budged for more than a year all while the x86 server component of it was growing at nice rates. Going back as far as three years this picture remains more or less the same (with ups and downs).
SGI isn't as detailed in its reports from what I heard Itanium based systems are the biggest revenue driver for them and their revenue is declining slowly. For them their growth in high performance systems clearly does not make up for the decline in workstations. This is for the period of a year. If you compare 2004 and 2002 you'll see that high performance systems and workstations both declined drastically (although the decline in workstations is even more pronounced).
So in general terms my statement that PA-Risc/Alpha/Itanium/R16000 market is not growing is probably justified. Itanium has not proven strong enough to grow this market significantly. Keep in mind that I am trying to be conservative with both these last statements for the reason that I have not investigated it thoroughly - it's from memory mainly plus some quick checks.
re: Convergence of x86 and EPIC platforms / cpu architectures.
BTW, I don't think the cpu architectures will converge either, but was wondering if you did.
re: IPF's biggest advantage over x86 is microarchitectural implementations that favours memory over complex logic and transistors over interconnect, both of which grow more important over time.
I think that's exactly why EPIC will be more cache dependant than x86. More cache means larger dies.
re: To date x86's biggest advantage over IPF is its software base and lower platform cost both of which shrink over time.
x86's software base will remain an extremely valuable asset. You're right about platform costs difference is shrinking. You don't mention though that x86 server cpu's will remain cheaper when compared to hypothetical EPIC cpu's of exactly the same die size because of economy of scale (production costs, development costs, etc...). Same goes for software prices to some extent. Provided that EPIC will never effectively emmulate x86 at a significantly smaller die size it won't stand a chance to reach similar economy of scale (alternative significantly better x86 emulation performance at same die size would do the trick too). Significant is 50% for example (big WAG), considering the large differences in economy of scale as they are today. We might never see such a situation, and if ever it won't be in the next couple of years for sure.
Regards,
Rink
