Replies to post #31616 on Advanced Micro Devices Inc (AMD)

[smooth2o]

Sgolds:So far I only see that the process hits a wall around 3GHz.

There is one other thing I think needs mentioning. Eventually, all major manufacturers need to surpass this barrier. AMD is no exception, and the concept does not involve rocket science. Intel is far, far ahead of AMD in knowing about how to solve this problem using both process and microarchitectural solutions. It could be one of the reasons that they are maintaining the netburst architecture.

Smooth

[dougSF30]

Remember the xBit rumor of a couple weeks ago? I think the 90nm problems extend beyond Prescott. And it's not valid to simply compare frequencies of Prescott and Dothan and conclude that because Intel can make 2.2GHz Prescotts, that implies that sub-2GHz Dothans are living up to Intel's TDP goals (21W was it?).

Looks to me like Intel's 90nm process is yielding parts with TDP's about 30% higher than the goal. I see no evidence that this will not continue...

Doug

[Dan3]

Re: Evidence?

So far I only see that

Dothan is 6 months late. Athlon 64 was 6 months late, too, but Opteron was early. Prescott was late and doesn't seem to really be here as an improvement over Northwood and Dothan is quite late.

Maybe they'll pull their 90nm process together the way AMD pulled their SOI process together.

But, so far, Intel's 90nm process doesn't look good.

[wbmw]

Sgolds, Re: So far I only see that the process hits a wall around 3GHz. Will they have any problems with a 2GHz processor?

How can you say something so silly? They have already demonstrated a 7GHz ALU, among other high frequency parts, on the same process. Frequency is not solely a process limitation. It is a design limitation based on the number of transistors in the longest string of combinational logic in any one of the pipeline stages times the time delay of each transistor. Obviously, small process improvements will reduce the time delay between the transistors slightly, and speed path fixes will reduce the number of transistors in the longest piece of combinational logic. Perhaps it's more accurate to say that the current stepping of Prescott is high volume up to 3GHz (although lower volume parts seem to have been released up to 3.4GHz). Dothan is a completely different CPU micro-architecture. There is no reason to believe at this point that Dothan will face the same clocking limitations of Prescott.