A post from Aces's - perhaps there is hope for us yet!
(I'm still holding to an expectation of $280 by the end of the year)
<VBG>
==========================================================
aceshardware.com
In Response To: Re: Copper interconnects -> some benefits ? (Martin McPherson)
> What is remarkable is that the Athlon (any version) can achieve the high
> clock speed it does with only a 10 stage pipeline as opposed to the
Integer pipe is 10 stages, FPU is 15 stages.
> Willamette's 20 stages for execution alone! According to JC and a chart of
> the Willamette's pipeline (2-24-2000), the Willamette's 20 stages do not
> include decoding! Its potential for ramping to high Mgz must be
> unbelievable! Perhaps 2-2.5 times the Athlon's, if clock speed is directly
> proportional to number of execution stages.
Clock speed is not directly proportional to number of stages. P6 has litle more stages in integer path (compared to Athlon) - It has 12(?) stages (10 incritical path, but there is also retirement and other cleanup stages) against Athlons 10. It's FPU pipe is shallower but it lacks all that sophisated allocation, renaming and such stages so after removing that complexity addictions buth are virtually the same. Now look at the clockability of Athlon against PIII. 700MHz on 0.25 was easy with K7, while P6 (old core, tweaked many times) maxed out at 600MHz in PIII incarnation, Intel has now serious problems even at 800MHz, while AMD claims no hard problems at least up to 1.1 GHz (on copper, though). So not only the number of stages is important but the design of the stages is very important too. Now look at even more dramatic example: K6-2 reached now 550MHz on 0.25 micron - thats only 50MHz behind PIII while K6-2 has 5-6 stages (depends on instructions) against PIII's 12-13 stages. Now Willy could go stably to 1.3 GHz, that 1.5GHz demo only booted up and ran application showing it's 1.5GHz - nothing more, thus CPU was probably severely (and unstably) overclocked. At release time they might have stable 1.5GHz but that's probably all (they will go to 1.6GHz shortly after with some tweaks, but for higher rate they'll need 0.13 micron. AMD showed (with K6) that thay are quite capable of tweaking up design and process to squeeze it to the max, so 1.4GHz Mustang (on 0.18 copper) is not out of question in Q4. Remember also, that AMD's Dresden FAB 30 has production equipment ready for 0.15 micron on copper, so in case of clockability problems thay should be able to relatively easily go half step onto 0.13 way making 0.15 copper parts (At that process 1.6 or 1.7 GHz Athlon should be possible).
Regards
Sebastian |
