Chuck, <could you elaborate on the gut feeling part of it.>
First of all, I originally felt that because the P6 core's pipeline was 12-14 stages, which is longer than Athlon's 10 stage pipeline, I felt that P6 should be just as scalable. Later, I found out that my assumption was slightly off, since even a shorter pipeline can be more scalable than a longer one if designed right.
Second, the P6 core has proven to scale rather nicely in frequency, even in comparison to the RISC competitors like Power PC, PA-RISC, and UltraSparc.
Third, I still can't shake my feelings that the Athlon core, as nicely designed as it is, still doesn't feature anything very revolutionary compared to what the P6 core already featured. It may be more scalable than P6, but not as much as people think.
Finally, with the exception of the two month Coppermine delay, Intel has always surprised on the upside when it comes to yields and binsplits.
These are the "gut feelings" that convinced me that P6 should be just as scalable as Athlon, or at least close. My sources, including PB, are merely confirming my feelings.
Tenchusatsu |
