[Info-vax] Did Ken Olsen kill Alpha?

Michael Kraemer M.Kraemer at gsi.de
Mon Oct 17 20:45:09 EDT 2011


Kenneth Fairfield schrieb:

> I think two issues often get lost when discussing what
> happened with Palmer and Alpha.  First, Palmer was head
> of the DEC fab(s) when Alpha was being developed, and
> when he took over from Olsen.  Second, fabs are *very* 
> *expensive* if you're trying to keep up with the
> technology curve: you really *must* have volume to keep
> them affordable, something Alpha never had in spite of
> Palmer's wet dreams about Windows...

at least some common sense in this discussion ...

> The cost of the fabs was probably the single biggest
> driver in the Intel lawsuit (IMHO). 

The initial cost was $500M, and they had to spent
another $200M annually, iirc, just to keep up.

> Palmer hoped he 
> could proceed with Alpha using the fabless model.

Even that was costly ($300M/a), according to

http://tinyurl.com/3o3mhrf

Putting these numbers together, I wonder how DEC could
survive that long. Even the quoted $1500 per chip
seems to be a dumping price.

> Unfortunately, while Intel and IBM do support fabbing
> other companies' products, they rarely do so on their
> latest technology.  That put Alpha at a permanent 
> technology disadvantage. 

When the last bunch of Alpha's were produced in IBM's fab
I think they used the latest technology (SOI,Copper),
the same as for POWER.

> (It's also much, much harder
> to debug a new shrink in the fabless model: you simply
> don't get the quick turn-around of fixes that you do
> if you own the fab and its scheduling.)

Maybe DEC were aware of this when they decided to
build an own fab?

> Since Intel is the big gorilla in terms of fabs, which
> they can afford because of their relatively huge volume,
> I'm actually surprised that IBM has been able to keep
> up.  I know IBM is big.  I know IBM sells premium priced
> high-end systems.  I'm still amazed that they're able
> to afford both the R&D and the actual construction costs
> of cutting-edge fab technology.

IBM always try to recycle their R&D output.
There are more Power designs than just POWER7,
and their fabs make a living from all kinds of
Power derivatives. For example, I think in 2009 they celebrated
50M PPCs for the Nintendo console alone, produced since 2006.
Add to that the chips for Xbox and Sony, plus embedded stuff.
That's what keeps the fabs alive.




More information about the Info-vax mailing list