Doug,
Wow, you must have some cool glasses that allow you to see the voltage and power dissipation of a part that has neither specified.
To the naked eye, the HSF did not look capable of cooling an extraordianry power dissipation. It is certainly high, but if I were to guess, it is not far beyond the the 3.0 GHz QC Intel is selling currently to Apple.
One of the reason I think that is that if you want to overvolt / overclock a part, you can't let the temps go too far up (because at higher temps, the clock speeds are unattainable). It would have required much bigger HSF to cool, say 200W, while keeping the temps reasonable.
I think the explanation that Faud suggested makes some sense, which is that only a very small percentage of parts are capable of clocking that high.
Do you have an alternative explanation for why AMD can't start selling these in a couple months, instead of Q208?
I think the new part is not a current stepping, but the next stepping that came out of some rocket lot, that hasn't gone through testing, and the earliest it can enter production is probably a September time frame.
That means that AMD will be stuck in the current anticipated clock speed range for the rest of 2007 - sucking wind in all single thread benches (outside of narrow heavy FP benches).
3 GHz Barcelona would have given Conroe run for its money, but it does not look as well against anticipated Penryn clock speeds (aided by substantial cache advantage).
The bottom line is that AMD is late again, has completely lost its momentum, and it does not look like AMD is going to regain its momentum any time soon. The best case scenario for maintaining some broad parity would be a 2.6 GHz in Q4, 3.0 GHz in Q1.
AMD is in process of blowing its only strong segment (4 socket servers) by dropping the ball on HT 3.0 for Barcelona. QC Barcelona would have had a chance to look much better in 4 socket servers if all 4 HT 3.0 links were used, to completely connect the processors with 3 links, 4th link for I/O. Bandwidth and latency would have been improved substantially - keeping up with the demand of 16 cores. As it looks now, HT 1.0 will be the bottleneck. Through the pathetic execution, AMD has turned one of its biggest assets to (close to) being a liability...
Joe |
