Rob, just to address a few points,
<21364 aka Arana aka EV8. Just discussed at MPF. Tench mentions 100% greater integer than 21264. That must be 21264 at its peak as EV8 is projected to be 200 SpecInt95 , 300 SpecFp95.>
I thought 21364 was EV7, not EV8. EV7 will integrate the RDRAM memory interface onto the chip. EV8 will introduce simultaneous multithreading. I don't recall what the schedules were, since I don't have my notes and my foils near me right now, but EV8 won't be out for quite a while.
<Intel has yet to release performance numbers for Itanium/Merced.>
The latest issue of MPF has a feature article on the Merced architecture. In the article, it says, "Intel claims a four-processor Merced system will outperform a four-way server with 1.1-GHz 21264 processors on transaction-processing benchmarks." It doesn't say who from Intel said this, and besides, it's not common practice for Intel to make such claims at this stage. But nevertheless, that claim is pretty amazing. (I don't recall hearing such a statement at the MPF conference, though.)
<So pipeline length and as the IBM fella points out "EPIC isn't high clock friendly" (paraphrasing).>
What do you think Marty Hopkins from IBM is going to say? It's his job to spread FUD. If EPIC isn't very accomodating to clock speeds, then what would his take be on the rumors that McKinley will have very high clock speeds? What if McKinley turns out to have higher clock speeds than IBM's own Power4?
Tenchusatsu |
