[Cialug] OT: More on Intel or Moron on Intel
dave at visionary.com
Fri Jun 10 12:07:42 CDT 2005
I know an engineer who has worked for Intel and is now at AMD. A lot of
what your friend says is similar to what I've heard from him.
IMHO - it would be really nifty to build an architecture where you could
plug & play multiple cpu cores to increase performance as needed. My
"dream" system would have an open bus architecture where cpu, memory,
storage, io and other stuff could just be plugged in as modules as
needed. If your computer isn't fast enough, you just go to BigBox
Computer Store and buy another processor module or two. You could also
have the system enable & disable modules as needed to reduce power & heat.
I'm sure some people with actual engineering experience can tell me many
reasons why this is just a dream... but I think it could happen eventually.
BTW - there are some Open Source processor designs in the works. Most of
the ones I've heard of are more to replace small device processors like
Nathan C. Smith wrote:
> Probably nothing new here for most of you, lots of opinions and guessing.
> I'm interested in alternate viewpoints and other opinions.
> One of our new attorneys (he is also an engineer) worked at Intel for a
> while, I had a brief conversation with him yesterday and some points I took
> away were this:
> * Intel has all the hallmarks of a big company including headcount politics
> * Chip foundries are expensive to build
> * In his opinion, Intel may no longer makes the "best" CPU chip. (I think he
> meant highest performing because they are certainly reliable)
> * There is more money to be made in volume chips for consumer electronics
> * Intel chip design teams are big.
> * Intel maintains a "skunk works" team
> * The manufacturing and design groups are in two "silos" that appear to be
> pitted against each other.
> He went into the usual schpiel about how the way to get more performance has
> to change because going smaller on-chip is getting to be extremely expensive
> (requires new specially built chip foundries) and physics are getting in the
> way. The complex Intel model of caches and tricks to boost performance
> breaks down when you go to a dual core chip, requiring other changes to get
> to the performance needed.
> Intel is also expanding into new areas including software and devices more
> aggressively than when they were tightly focused on chips. Blame or thank
> computer saturation and dot-bomb for that.
> What I think it boils down to, is that is is getting too expensive to build
> "hot rod" chips, the way it has been done, and Intel is finding other ways
> to get the performance without building more expensive chip foundries. He
> said one of the last foundries cost almost 10 billion and that is _before_
> the chip-making equipment went in, basically the building, HVAC (clean
> rooms), and the plumbing (water purification, toxic material disposal etc).
> A large company doesn't change 20 years of momentum in a day or even a year,
> but I think at Intel commodity chip-building is going to change from
> building expensive, fast chips to doing more with less until a quantum leap
> occurs that takes us to a new way of thinking about chips.
> OK, he's not the ultimate insider, but he was at Intel and he has first-hand
> impressions of what goes on. Its easy to forget that so much crap and
> politics go on at a big company like Intel that it isn't always about
> building the best chip. It's about budgets and shareholders and dollars.
> [WOT] I wonder what a free (as in libre) chip would be like? GNUchip? You
> can only sell the packaging it comes in.
> Cialug mailing list
> Cialug at cialug.org
More information about the Cialug