Re: I get clovertown 2 socket specfp rate at 104, versus opteron 2 socket specfp rate at 92.1...
Pete is referring to a submission from Sun that uses additional optimizations and compiler tricks which have so far eluded the other vendors. But be that as it may, Opteron does have a higher SPECfp_rate score in that particular submission (119 using the peak measurement, and thanks in no small part to the performance of 179.art).
Of course, most educated readers who understand the SPEC benchmark know that the performance is restricted to the environment in which it is run; that is, using Sun's platform under their OS using their latest compiler techniques on code that has been studied long and hard to squeeze every last drop of performance.
And taking into account that Sun has a small sliver of the market, that only a percentage of Sun's platforms ship with Solaris, that most applications haven't been built using Sun's latest compilers, and absolutely no application has been studied and optimized to the same degree as SPEC_CPU, then you are left with a pretty contrived situation.
Of course, Pete knows this, but he's hoping that you don't.
Servers aren't bought to run SPECfp_rate, and they're not even bought to run workloads that even resemble SPECfp_rate. I've been arguing this endlessly, but to no avail. The people interested in touting Opteron's SPECfp_rate performance are generally not interested in putting it in the appropriate context.
Then again, this is how it's been on these forums for years now. You've seen the data, and you know how Clovertown performs in other benchmarks. Hence, you can come to your own conclusions.
Market Data 
Markets 
AI
