DEC veterans prepare chip challenge for Intel, AMD, IBM and S un

[Bryan J. Smith]

"Robert M. Riches Jr." <rm.riches at verizon.net> wrote:
> I can't speak for any difference between the Power and
> PowerPC instruction sets,

They aren't compatible.

Research the PowerPC 620 design failure, and the split IBM
made on PowerPC (from Apple-Moto) in the Power series for
several years, until the very recent PowerPC 970
(Power4-based).

> but I can offer a perspective on chip power consumption
> vs. performance.

Power comsumption v. performance is a powerful metric ... for
_embedded_ systems.  For desktop systems, consumers seem to
be more interested in performance v. dollar-cost.  That's the
difference.  ;->

Furthermore, dollar-cost is all about economies-of-scale.

That's why Apple is going Intel.  Intel can deliver many
different x86 products for many different Apple products,
because they are in use by many different solutions already. 
Apple is merely becoming another Dell for Intel, not their
sole partner, just preferred.

With IBM (or Moto for that matter), Apple was basically the
sole customer (outside of itself) for several of its
products.  In fact, it was the number of products IBM had to
roll out just for Apple that was part of the problem.

IBM wants partners who ask for a spin of *1* product in their
foundary, and in high quantity (millons).  That's Sony. 
That's Microsoft.  That's this company, and many others. 
These are "fixed configuration" solutions that are run for
years.

Apple expects IBM to "keep up" with revs, release different
speeds, peripherals, etc... in quantities of only hundreds of
thousands.  That's not what IBM was interested in, and why
IBM couldn't deliver.  But Intel can, because it sells
millions of these different products already -- at the
leading edge -- giving both performance and price.

Power is secondary in many cases.

> The PA Semi story was all over the financial headlines for
> related stocks symbols yesterday (Oct. 24).  While the chip
> is supposed to be much lower power than conventional chips,
> it is supposed to have very good performance.

Of course!

In fact, the original promise of the PowerPC 970 was that it
was supposed consume 11W at 1.4GHz, and less than 20W at
2.0GHz.  That was back in 2003.  But where are the Apple G5
notebooks?

Again, goes back to the economies-of-scale issue.  Price is
more of an issue than power for most end-consumers, even for
most notebook sales.  The fact that IBM had trouble
delivering many different products in quantity and cost --
especially so Apple could release a notebook G5 -- was the
ultimate issue.

Something Intel will not have an issue with doing, and can
deliver far cheaper -- especially when it comes to portables.

How is this product going to do at a leading-edge product?
What kind of performance will it deliver in 2007 in
comparison?  Is the power savings going to overcome the
performance and economies-of-scale that Intel can deliver?

And, lastly, it's still not PowerPC compatible.  ;->

> For a given FAB process and such, if you graph performance
> as a function of power consumption, the curve is strongly
> concave up.  On the end denoting high power and high
> performance, the curve is extremely steep, nowhere near
> linear from the lower part of the curve.  Put another way,
> the last 90% of power consumption may only give 20% extra
> performance.

On the _same_ architecture.

But if that was the case, again, why didn't IBM crank out the
lower-power PowerPC 970s?  The answer is, once again, the
reality that IBM can't deliver so many different products
that are short-lived (9-18 months) and make it worth their
while in only hundreds of thousands of units -- at least not
compared to Intel.

Now let's look at this company.  Throw out the PowerPC issue
for a moment.  Is there any reason this company will be able
to customize 3-4 different products for just Apple any more
than IBM?

Or it is more likely they will run millions of "fixed"
configuration products for a 3+ year product that is
unchanged, much like IBM has done for Sony, Microsoft and
others?  Things that are not exactly "leading-edge" and
"equivalent performing"?  Especially when the company has to
rev new products every 9-18 months?

> A good macro-architecture, a good micro-architecture, and
> a brilliant design team can gain back most of that 20%
> performance while keeping most of the 90% power reduction.

I'm not disputing this.  And this is the reality of
_embedded_ products.  It wouldn't surprise me if I'd find one
in a souped-up iPod come 2007.

But what does it do to address the performance of a typical
desktop?

> Yahoo has a story that says the processor will have two
> 2-GHz cores.  That ought to be fairly respectable.

Fairly respectible doesn't sell to 90% of PC consumers.  ;->

Add in the fact that Apple wants 3-4 products rev'd every
9-18 months and _no_one_ is going to give them a performance
v. price like Intel.

> I wouldn't be surprised to see a chip with acceptable
> performance for at least desktop use, definitely acceptable
> for laptops, and maybe good enough for some servers.  For
> servers, don't forget that supplying power to and removing
> heat from a few dozen high-density server racks packed into
> a small machine room can cost a lot of money.

And maybe IBM itself might consider the design for its
Linux/Power solutions, possible a new Blade series.

I know others definitely will!

But what does this do for Apple?

> If you can get 80% of normal performance and only use 40%
> of normal power,

On the _same_ architecture.

You're not looking at comparing to a _different_ architecture
where the lower envelope is 25-50W, but it's already
delivering 3-5x the performance of a product designed for
8-13W.

> you can use 25% more servers and save a lot of money
> in power and cooling.  (I heard of a special deal in which
> a large search engine company bought a huge number of
> special reduced-power processors from a large semiconductor
> maker, and power reduction was the primary goal.)

I don't dispute this one Iota.

I only dispute how this would affect Apple's move?

> (For credibility evaluation, my previous job was overseeing
> the technical side of schematic formal verification of the
> Pentium 4 processor.  I had been on the design teams of
> three other processors before that.)

I know.  We've spoken before.  ;->


-- 
Bryan J. Smith                | Sent from Yahoo Mail
mailto:b.j.smith at ieee.org     |  (please excuse any
http://thebs413.blogspot.com/ |   missing headers)