chipguy,
1) Too bad PC myopic people think that SPECint accurately
reflects the commercial or technical workloads that most
people buy large scale servers to run.
Who said accurately? Spec benchmarks are relatively easy and inexpensive to run, the results are readily available, so all one has to do is to pick which correlates with your app better, Int or FP. SpecInt correlates well with majority of the commercial apps, server and client oriented, including the most commercial server app - database servers.
If that was the case Alpha would have been much more successful than SPARC instead of the other way around.
There is more to commercial success than superior benchmark scores.
2) Who says that on equal process footing that P-M will be
better on SPECint than IPF? At 1.5 GHz the 130 nm Banias
gets 995 SPECint_base2k compared to 1408 for the 130
nm Madison 6M. IPF's advantage in SPECint IPC should
increase in 90 nm with Montecito vs Dothan as well as the
IPF chip likely having better frequency scaling at 90 nm.
You are comparing chips that may or may not be released next year with chips available for sale today.
Extend P-M to server class features and you seriously
close the gap in cost and power vs IPF that makes the
Dothan so intriguing as a fast cool little chip.
I have hard time imagining the power going up 3x. As far as cost (silicon cost), again, Dothan has somewhere around 3x advantage today.
Joe
Market Data 
Markets 
AI
