Pete, <Williamette will be slower clock for clock in high precision fp than Athlon.>
I think everyone, especially the AMDroids, is underestimating Willamette's FP performance. And that's without using SSE2 instructions. Intel didn't give too many specifics, which is why people are assuming the worst. But people I've talked to in the Willamette team are very confident about its floating-point performance. I doubt these engineers would place such confidence in "smoke-n-mirrors."
<With one very broken pipeline against three high performance pipelines, it is acknowledged that Athlon has more fpu power but, possibly, ISSE2 may be able to even score.>
I've yet to see that oh-so-great FPU power. Certainly Athlon has the potential, given its strong FPU. Yet even Thunderbird failed to blow Coppermine out of the water in the benchmarks, despite the so-called "three high performance pipelines." And let's not even mention SPECfp, thanks to the lack of good Athlon-optimized compilers.
<The Mustang fpu register flat file will improve the fp capabilities to RISC like levels leaving Willie in the dust. Sledgehammer, its 64 bit extensions, and dual core will probably exceed HP's PA-RISC at the time and since HP says even McKinley will not touch it's RISC, McKinley is toast.>
First of all, it's Sledgehammer which will get the flat FPU register file, not Mustang. And Sledgehammer is slated for the end of 2001.
Second, Sledgehammer will start off as a single-core processor. It doesn't make any sense to introduce a brand new architecture in dual-core form from the beginning.
Third, not even AMD expects Sledgehammer to compete against IA-64 in pure performance, which is why AMD is sidestepping the issue. "We will not relegate 32-bit performance to 2nd class status," claims AMD, but they stay silent on how well their 64-bit extensions will actually perform.
And fourth, HP is planning on moving its entire server line to IA-64, starting with Itanium, and continuing with McKinley. If McKinley "won't even touch PA-RISC," why is HP making the IA-64 jump in the first place? Heck, why would HP even work on McKinley if PA-RISC will beat it?
<There already is some K75s that have reached 1.1GHz (see Moldyn, QMC scores) and 1.4GHz (133 x 10.5) Tbirds are demoed awaiting, per rumour, only for 133MHz motherboards which, Ali says could arrive August or September this year.>
I'll tell you what I told the PowerPC advocates seven years ago ... "I'll believe it when I see it."
Tenchusatsu |
