To: Tony Viola who wrote (75721 ) 10/17/1999 3:04:00 PM From: Ali Chen Read Replies (4) | Respond to of 1572449
Tony, <The benefits of Cu at .18 have been over hyped, IMHO. Al plainly still works at .18...> Maybe still works, so what? AMD has decided to polish the Cu technology on less demanding 0.18 process, so their transition to 0.13 apparently will take much less efforts (given experience obtained on 0.18). If Intel decided to start with 0.13, good luck <ggg>. They would need to solve SIMULTANEOUSLY new lithography problems with copper problems. It is clear that AMD was much smarter at this point. Intel seems to miss one more inflection point. <Why do you deny the wisdom of Intel's opinion in areas (process and manufacturing engineering) where they clearly have been superior to AMD forever?> Wisdom? You are sadly mistaken here, as usual on every technical matters. Contratry to public monikers, Intel has NEVER been superior in process and manufacturing engineering. Am386 processors were driven to 40MHz while Intel dropped i386 at 33MHz. Lead = 20% in process tecnology alone. Am486 processors were manufactured even more superior to Intels: i486 stopped at 100MHz, AMD made them up to 150MHz, or 50% better in terms of FETS quality (very simplifying). If you want to refer to K6 "problems", the conclusion about AMD manufacturing deficiencies is also false. The fallacy mainly comes from the above two examples, where logical design of CPUs was essentially IDENTICAL. This is very different with K6 versus P6. I'll try to teach you again. Some basics first: 1. K6 design has the pipeline of 4-5 stages long. In contrast, P6 core has 10-12 stages, or 2.5-3 times longer. Now remember, both processors have to execute same instructions, therefore the amount of logic is presumably the same. But the difference is that for K6 all this logic is divided between 4 stages, while for P6 is is spread out over 12 stages. This makes each stage much shorter in terms of CMOS gates, and therefore in signal propagation time, which in turn determines the top working frequency of the CPU. 2. From the above "microarchitecture thingy" we can conclude that the K6 has 2.5 times more gates between clock ticks than P6. Now the task for you: if AMD K6 design topped at 500MHz, and P6 is topped at 600MHz, HOW MUCH BETTER the AMD FETS must be to achieve this? The answer: 2.5 *500 /600 = 2.08 ! Therefore, AMD "transistors" must be 100% better that those of Intel to achieve this "parity"!!! Please note that this "superpipelined" design was stolen from Alpha, and Nexgen nor AMD had no access to this advanced design technology. Intel just milked out every advantage from this superpipelined architecture, with very "concervative" process. Now, when AMD eventually has even better microarchitecture with K7, you can estimate the effect of AMD's 100% lead in FET technology on CPU frequency by yourself. Have fun.