Re: EV79 CANCELED !!!!!!!!!

From: Robert Deininger (rdeininger_at_mindspringdot.com)
Date: 10/25/03 

Date: Sat, 25 Oct 2003 00:12:29 GMT

In article <3F98AA90.E184BCC8@istop.com>, JF Mezei
<jfmezei.spamnot@istop.com> wrote:

>Lets not forget the rumours that EV7 was ready well before it was actually
>released, with much speculation that HP didn't want to release EV7 before IA64
>so that IA64 could have its 15 seconds of fame before it was outperformed by
>Alpha.

You ought to forget those rumors (didn't you start them?), since they are
pure, undiluted bull***.

The Marvel systems went out the door with the best EV7 CPUs that could be
produced at the time. The chip and systems engineers spent most of the
previous year finding and fixing serious problems. A couple of months
before ship, it was far from clear that there would not be further
delays. Another revision of the CPU chip might have been needed, and
would have meant months of delay. Fortunately, the bugs were fixed or
avoided the systems were released.

>And then, when EV7 was allowed out, its performance was less than had been
>anticipated.

How much longer would customers have waited for systems? They got the
best available at the time. Holding back until now would have allowed the
first systems to be somewhat faster. Waiting until next year would have
made them even faster. Would that be good business for HP?

 
>> The OpenVMS port to Itanium has continued and is on
>> schedule.
>
>Very few doubt this.
>
>> as EV79. As the project progressed, it became clear in working
>> closely with our CPU chip supplier that the EV79 chip would not meet
>> our expectations for performance or time to market.
>
>If their expectations were that IA64 would outperform Alpha, then of course,
>the EV79 wouldn't meet their expectations.

After several passes of EV79 prototype chips, they did not _work_ at
anything approaching the design speed. They were absolutely not ready to
ship. Additional prototype passes would have been required, and it was
virtually certain that nothing could have been given to customers until
post-2004. The problems were largely out of HP's control. Throwing tons
of money at it wouldn't have fixed the schedule problem, at least not
enough to matter.

>> However, the manufacturing process maturity for EV7 has progressed
>> very well, producing good yield and making it possible to produce an
>> EV7 part (EV7z at 1.33GHz) with significantly improved performance
>> over our current EV7.
>
>So, what this means is that EV7 won't even need any changes, they'll just use
>a faster clock.

You think that's easy? Why don't you start up a CPU company?

>And it may also mean that EV7 was purposefully given a slower
>clock, and now, instead of doing EV79, they are just going to remove the speed
>restriction they had put on EV7.

There was never a "speed restriction" on EV7, except the requirements that
it could be fabbed in useful quantities and that it work reliably in
systems.

>Sorry of I appear sceptical, but it was up to Comapq and HP to work hard to
>regain trust, which they didn't.
>
>> Therefore, we have made the decision to provide our customers with the
>> performance improvements that we had committed based on EV7z rather
>> than EV79.

>In the end, HP breaks the "roadmap", "commitments" etc. They promised EV79
>which was a process shink. And they will not honour that promise.

EV79 was NOT a "process shrink", whatever that vague term might mean.
Bleeding-edge CPU fabbing is more art than science, and changing ANYTHING
in the design invites unexpected problems in the real chips when they
arrive. It's a very long distance from a CPU that works perfectly in the
simulator to one that works in silcon. EV79 had real, algorithmic changes
in the functional parts.

Other verndors are not immune to this. Why do you think Intel keeps 3 or
4 CPU design programs running in parallel? If (when) one has problems,
the others still have a chance to keep new chips coming out.