Replies to post #33810 on Advanced Micro Devices Inc (AMD)

[CombJelly]

"My question to all is, what is keeping Intel from using Nocona in 4P and up servers?"

Probably not a whole lot, but Nocona doesn't have a L3 cache. Given the shared bus topology, that would put a lid on performance. That might be the plan, assuming Potomac gets the knife. But then it's going to look even worse against 4 and 8 way Opteron....

[HailMary]

What am I missing?

I'll speculate an answer to my own question.

Dothan could already have x86-64 in it. That would give Intel the bridge they need to get something out in 2005 that can compete with Opteron. I really think demand for x86-64 for servers/workstations is going to be huge in 2005, so much so that it could be a banner year for AMD if they don't have competition in that space.

[sgolds]

HailMary, when do you think that Intel understood internally that 90nm NetBurst was broken? Then, when did they realize it is not fixable?

A company like Intel will start a skunkworks at the first point where they realize there is a major problem. This is the point they would have asked (some individuals on) their Israeli team to quietly work with (some individual on) the Oregon team to transfer technology and evaluate the potential of Dothan to be a foundation for the x86 family.

When they reached the second point then they would have started the reorg to put full resources on the project. According to today's articles, the reorg is history and all hands are on board the new strategy.

The last thing a company does is announce it to the world. That is rumored to be today?

Here is my speculation on dates: Just about a year ago Intel would have noticed major problems on 90nm Prescott, but the engineers (who, like everyone, emotionally defend their work) would have insisted on time to fix the problem. So the skunkworks was probably started just about a year ago.

There were delays in Prescott late last year. The Prescott team had one last chance then to make it work. They couldn't. Textbook management then says to productize anything completed, even if it is inferior, cancel the follow-on products and reorganize. That appears to be what happened.

So I think the reorg was formulated late last year and implemented in stages right after New Years. By that time the skunkworks probably had specifications for an AMD64 Dothan and a dual-core version of it. Project teams would have received their marching orders, including a detailed specification, probably in January. Prescott was in manufacturing and developmental Engineering was available for the new assignment.

Starting from a good specification (vs. idea stage) means that it should take about 12 months to move a product to the prototype stage and introduce it into manufacturing. Give it another 6 months of refinement and scaling to volume, that gives new product in mid 2005.

[Petz]

HailMary, a lot of good questions. I think the shortest route to a dual core Dothan would be an MCP package with 2 cores, a big L3 and a bus arbitration logic chip.

Petz