Re: Why did VMS users go along with the itanium farce?
From: Bill Todd (billtodd_at_metrocast.net)
Date: 09/30/05
- Previous message: toby: "Re: Pete My Boy...yer a Genius"
- In reply to: Main, Kerry: "RE: Why did VMS users go along with the itanium farce?"
- Next in thread: AEF: "Re: Why did VMS users go along with the itanium farce?"
- Reply: AEF: "Re: Why did VMS users go along with the itanium farce?"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Fri, 30 Sep 2005 00:34:33 -0400
Main, Kerry wrote:
>>-----Original Message-----
>>From: JF Mezei [mailto:jfmezei.spamnot@teksavvy.com]
>>Sent: September 29, 2005 3:03 PM
>>To: Info-VAX@Mvb.Saic.Com
>>Subject: Re: Why did VMS users go along with the itanium farce?
>>
>>"Main, Kerry" wrote:
>>
>>>1. Although it would be nice, Itanium does not need to be
>>
>>the leader in
>>
>>>latest speeds-n-feeds in order to be successful. It just
>>
>>needs to be
>>
>>>competitive. Heck, Sun has proven that over the years and as slow as
>>>SPARC was,
>>
>>
>>There is a big difference here. SPARC was succesfull, considered the
>>"industry standard" for Unix and was widely regarded as a
>>safe platform.
>>
>>IA64 has none of those attributes. It doesn't have an established user
>>base, it is new and already talk of killing it because it just isn't
>>what it promised it would be. And more importantly, HP and Intel have
>>already started to cannabalise it by limiting its market niche to
>>supercomputers, and then there is the writing on the wall with the 64
>>bit 8086 which Intel swore woudl never happen, coming to
>>further reduce
>>IA64's remaining niche.
>>
>
>
> Wow, where have you been getting this stuff?
I'll answer that question just to keep you honest, Kerry - though that's
always a difficult task.
"There is a big difference here" indeed, and he goes on to explain it,
point by point:
"SPARC was successful" - very much so, to the point that it established
a loyal user base large enough to have weathered years of sub-par
performance and still remain the largest single Unix customer base.
"considered the 'industry standard' for Unix" - absolutely. AIX made
headway much more because it was IBM's entry than anything else, and
PA_RISC never enjoyed nearly the perception of being the standard that
SPARC did (and Tru64/Alpha, good though the combination was, remained a
distant 4th, though was closing with pretty impressive speed when it got
the axe).
"was widely regarded as a safe platform" - yup, that's what going with
the leader gets you.
"IA64 has none of these attributes" indeed, and he lists why in detail:
"It doesn't have an established user base, it is new" - well, duh:
that's the penalty you pay when you introduce a new platform, so no
possible argument there.
"and already talk of killing it because it just isn't what it promised
it would be" - let's address this in more detail, lest you try to weasel
your way out of it:
1. Merced was originally slated to be the platform that left all
competition in its dust (2x - 3x the performance of the competition, and
the infamous Albert Yu quote that said competition had no future), then
turned out to be a complete joke (even Intel warned people ahead of time
that it wasn't anything more than a familiarization mule).
2. As late as early 2002 HP spokespeople were predicting a meteoric
rise in McKinley sales to levels that Itanic *still* hasn't matched 3
years later.
3. McKinley clock rates were scaled back from 1.4 GHz to 1.2 GHz to
eventually 1.0 GHz - and some parts wouldn't even run at over 800 MHz.
4. HP used carefully-selected benchmarks that continued to give the
gullible a far better picture of McKinley's relative performance than
was actually the case.
5. Madison clock rates were less over hyped than McKinley's were, but
still failed to rise over time to the levels eventually hoped for.
6. And now Montecito has just slipped another 6 months and when it
finally does appear won't even clock as fast as current Madison IIs
(with Montvale now targeting the frequencies originally expected for
Montecito and nothing more in sight before 2008).
So Itanic has beyond any shadow of a doubt failed to come anywhere near
what it was promised to be - time and time again with what has become
grinding predictability. As for the "talk of killing it" - well, that
would hardly be surprising under the circumstances, but there's also
well over a year's worth of Intel and HP behavior to lend credence to
that 'talk':
1. Intel killed its own new Itanic chipset work in the spring of last
year, leaving Itanic dependent on the aging existing chipset until
Tukwila (slated for 2007 then, now looking a lot more like 2008) save
for any chipsets developed by OEMs: Fujitsu and Hitachi already have
their own new chipsets, though they haven't posted any impressive
benchmark results with them yet; HP has one on the way as well - but
Dell, of course, does not.
2. Intel pulled the ex-Alpha team off their Tukwila core work around
December, 2004, and instead decided to make Tukwila yet another spin of
the already-tiring McKinley core used in McKinley, Madison, Madison II,
Montecito, and Montvale.
3. Intel scaled Montvale back from a significant revamp of Montecito in
65 nm. to a barely-perceptible upgrade in the same 90 nm. process (in
fact, as noted above, to just about the low end of what Montecito had
been expected to provide in the first place).
4. HP embraced Opteron with somewhat surprising enthusiasm considering
its potential for disrupting entry-level Itanic sales, and continued
beating the 'industry-standard' drum not on behalf of Itanic but on
behalf of 'non-proprietary' platforms based on Windows and Linux
(neither of which of course requires Itanic in any way). While this is
certainly consistent with HP's apparent desire to become a box-assembler
using other people's technology, it's hardly the behavior of a
corporation that thinks it has 'bet the company' on Itanic and thus
needs to bolster that platform preferentially.
5. And of course there are the moderately credible rumors that Intel
has approached both SGI and HP about the possibility of supporting Xeons
as well as Itanics in their large systems (presumably when the common
system infrastructure appears in 2007 - for Xeon, anyway).
So, yes - with the repeated cut-backs in Intel's support for Itanic
coupled with an apparent lack of urgency on the HP end of things and the
continuing dismal comparisons of Itanic reality vs. Itanic predictions,
one can hardly dismiss 'talk of killing' it as pure FUD.
"HP and Intel have already started to cannabalise it by limiting its
market niche to supercomputers" - yes, JF misspoke, and you wisely
pounced on that to quibble about. Had he correctly characterized the
niche as high-end servers, you'd have had nothing (not that this usually
slows you down, of course: you'll take spin over substance any day,
though often do make the effort to distort at least a pinch of substance
to add a thin veneer of lipstick to the resulting pig).
"and then there is the writing on the wall with the 64 bit 8086 which
Intel swore woudl never happen, coming to further reduce IA64's
remaining niche" - interesting development, that. Less than two years
ago Intel was indeed still pooh-poohing the idea of 64-bit x86, even
after already having built the capability into its own products just in
case. But when the writing on the wall became clear, they didn't
hesitate to reverse course, meet AMD head-to-head, and stop trying to
protect Itanic's ability to establish a foothold in that area and
instead throw it to the wolves.
I'm sure there's a lot more detail that I could come up with if I
considered you worth the effort of doing some actual digging, but that
will just have to do for now.
Have a nice day,
- bill
- Previous message: toby: "Re: Pete My Boy...yer a Genius"
- In reply to: Main, Kerry: "RE: Why did VMS users go along with the itanium farce?"
- Next in thread: AEF: "Re: Why did VMS users go along with the itanium farce?"
- Reply: AEF: "Re: Why did VMS users go along with the itanium farce?"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
|
