News Focus
News Focus
icon url

tatertot

11/14/09 2:50 AM

#85336 RE: fastpathguru #85331

Will AMD need to spend billions every few years to build new fabs and keep them on the bleeding edge?

In a sense, yes. They just pay per wafer, instead of per fab, now. And to the extent GloFo gets more customers, the process is not tuned solely to AMD's requirements any longer.
icon url

tecate

11/14/09 8:53 AM

#85339 RE: fastpathguru #85331

Feel free to disagree, but IBM's poor yields are legend, even at IBM.
icon url

Elmer Phud

11/14/09 9:16 AM

#85342 RE: fastpathguru #85331

There has never been an x86 manufacturer of AMD's caliber running on a fabless model.

AMD is not an x86 manufacturer. They are an x86 designer who needs to use a foundry with a process designed by a committee. They will rely on not terribly friendly external production teams who are already supporting multiple other products. If they choose to use multiple foundries they will have multiple processes using multiple tools, models, libraries etc supported by multiple competing production teams. Their products will underperform their potential because of the generic process, not their design deficiencies. If one fab has a problem they will be on their own with nobody to support them. Production running at one foundry can not be diverted to another foundry in a crisis. If it's designed for foundry A it's stuck there.

An integrated design/manufacturing house works as a team. The design and process are developed for each other, not in spite of each other. They can be adapted to the product's needs to enhance yield or performance. There are dedicated teams to support each product. All fabs run an identical process and any fab seeing issues has the support of the entire organization. You think GloFlo is going to help TSMC solve their yield problems? Any product can run at any fab. They are not competitors, they are team members.

If you think for some reason that AMD is somehow comparable to Transmeta, etc., well, I just... disagree.

Of course you disagree. You know nothing about this business and you believe AMD's PP slides. Had you been following AMD for 20+ years you'd learn to put AMD's promises in their proper perspective.
icon url

chipguy

11/14/09 1:18 PM

#85351 RE: fastpathguru #85331

There has never been an x86 manufacturer of AMD's caliber running on a fabless model.

That is because AMD burned its furniture and sold off the
family silver to stay in the manufacturing game as long
as it possibly could. Why? Because there is significant
advantage for both product margin and design quality for
high end processors doing it all (device design, process
design, and manufacturing) under one roof. The beginning
of the end of Alpha supremacy was when DEC divested
its Hudson fab to Intel and went on to a foundry model. It
never again had the huge performance and technology
lead ADT had vs everyone else in the early to mid 90s.

It is all pretty simple actually. When manufacturing at
AMD was spun off into a seperate corporate entity, GF,
a bunch of things happened. First of all a new level of
corporate overhead, SG&A, springs up at GF which has
to be paid for from the per wafer price AMD pays to GF.
Secondly, all discussion between AMD designers and GF
process guys are now inter-corporate and there has to
NDAs signed and so on. It's no longer a case of simply
walking down the hallway for a casual chat. Thirdly the
guys at GF care about GF succeeding and that means
attracting new customers and non-x86 business. AMD
is the legacy customer that pays the bills for now but
anyone who has ever worked in the tech industry knows
how much attention legacy business gets compared to
the effort that goes into pushing into new areas and
gaining new customers. This will really kick in once
AMD divests the rest of its GF holdings to pay its bills.

It is amazing to me that anyone thinks that going fabless
will improve AMD's ability to hold let alone expand the
~17% of the x86 processor market it has now in direct
competition with the product design/process R&D/manu-
facturing moster that is Intel and make the profits that
will allow it to survive in the long run. I wonder how many
families in economic distress think it makes sense to
sell off their kitchen appliances and eat at restaurants
full time. :-D