Re: Portents of Itanium death

From: Bill Todd (billtodd_at_metrocast.net)
Date: 06/11/03 

Date: Wed, 11 Jun 2003 01:08:09 -0400

"Rob Young" <young_r@encompasserve.org> wrote in message
news:uPL9xGJDnIjR@eisner.encompasserve.org...

...

> Intel has big speed bumps ahead with Xeon MP. Should keep Xeon
> competetive with Opteron, especially in those pesky tpmCs.

You actually got it right above: Xeon *will* likely remain competitive with
Opteron (save, of course, for being limited to 32-bit use). But then below
you, as usual, got carried away by your own spin.

> For example, this 4-way Dell config gives:
>
> http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=103042201
>
> Exec summary:
>
> http://www.tpc.org/results/individual_results/Dell/dell_6600_210403_es.pdf
>
> 78000 tpmC. The configuration for that is 4 - 2.0 GHz with 2 MB
> L2 with each CPU. As it just so happens, Intel is set to
> ship at month-end an upgrade of Xeon MP:
>
> http://www.theinquirer.net/?article=9823
>
> "As we've written earlier, the Xeon MP 2.8GHz with 2MB cache will be
introduced
> at the end of this month at around $3,700."
>
> Now it isn't rocket science to presume that a 40% increase in
> CPU speed will translate in quite a bit more tpmC.

Indeed it should.

>
> Here is an observation.
>
> AMD launches Opteron in April, posts tpmC to tpc.org for a
> 4 processor box that you can take delivery of in July. Quite
> good numbers - to look at, but you won't be running that config
> for a while.

Sort of like Madison posting 4-processor TPC-C numbers in April for a system
with an August 1 availability date, I'd say. Even if it's the same Madison
system that you'll be able to buy at the end of this month, the difference
between the two comparisons is only a couple of weeks - hardly enough to
suggest that AMD was being a tease and HP was not. And when it came to the
SuperDome results, HP posted TPC-C results in May for availability in late
October...

  Meanwhile, as part of a true
> launch campaign with savvy marketing Intel is set to launch
> a CPU when Opteron's become available that knock Opteron around
> on tpmC. The Intel boxes will be orderable. Why tease?

Let's see: only a day or two ago you were describing how AMD was remiss for
*not* having posted 2 GHz Opteron benchmark results yet, and now you're
praising Intel for precisely the same behavior? You're really living up to
dumb-blonde cheerleader stereotypes here, Rob: we'll have to get you
pom-poms and a cute little outfit.

>
> AMD does good on paper, Intel delivers in reality.

Intel has been delivering Itanic on paper for close to a decade now, Rob -
certainly since the original late 1997 projected release date (though that
date then itself got projected, and projected, and projected until the
product earned the nickname 'unobtanium'). And even as of today Intel has
yet to deliver an Itanic that customers actually want to buy.

  Wall Street
> sees these things too.
>
> Just as significant, back to the speed bumps for Xeon. In the
> next year Intel is set to take Xeon to 3.6 GHz with an 800 MHz
> front-side bus and 4 MB on-chip cache.

But not, I suspect that you'll find if you study their plans in a bit more
detail, all in the same chip.

  While others are quick to
> point out Opteron will be at 2.6 GHz whenever, you can see that AMD
> will be quickly falling behind the speed curve.

I guess one might think that if one were as ignorant of AMD's Opteron
schedule as you seem to be. 2 GHz is planned for this summer, 2.2 GHz for
late summer/early fall, and at least 2.4 GHz before year's end (when the *1
MB* cache Xeon is planned to hit 3.2 GHz, last I knew).

By a year from now Opteron will be in 90 nm (just like Xeon) and heading
toward 3 GHz and beyond. And it's likely to be either dual-core or have its
cache enlarged to 2 MB by then.

>
> Today, a 1.8 GHz 4-CPU Opteron does 80000 tpmC.

Not to be picky, but it's over 82,000 tpmC, Rob - and without the TPC-C
tuning expertise of HP behind that result.

  Today a 2.0 GHz
> 4-CPU Xeon CPU box does 78000 tpmC. Next year, Opteron will be
> pushing 2.6 GHz, Xeon will be at 3.6 GHz. Do the math.

Pentium will likely be at 3.6 GHz by then, and perhaps even the lower-end
Xeons, but the large-cache Xeons have so far always lagged *significantly*
in their clock-rate increases. So the math doesn't quite work out the way
you'd like it to: Opteron should itself have a bigger cache (unless it goes
dual-core instead), should be pushing 3 GHz by then, and will in all
probability still easily match both the slower, large-cache Xeons and the
faster, smaller-cache Xeons - while offering the distinct advantage of
64-bit operation for server environments that benefit from it.

>
> > We keep being told how great IA64 *will* be. But that usually means that
> > today, the chip is crap and not a success. The day they announced the
murder
> > of Alpha, it was TODAY that counted, not some time in the future.
>
> It will be great.

Not before 2006 at the earliest. Before then it's stuck with the McKinley
core, and will be nothing more than adequate (and a pig for power to boot).

> Intel is spending the money to ensure that.

Bill Gates doesn't have enough money to make that happen any earlier than
2006: read The Mythical Man-Month.

> Don't discount Xeon either. AMD has to fight a multi-pronged
> battle.

And is ideally positioned to do so - better than at any time in the past.
In competing with IA32 it has the 64-bit option to offer as an unmatched
feature (plus 100% IA32 compatibility so it's not any kind of a trade-off).
In competing with Itanic (assuming Itanic turns into something one might
care about competing with) it has drastically lower prices (both for the
processor itself and for the associated boards, due both to higher volumes
and lack of need for external glue) - plus, again, that great IA32
performance for mixed environments.

  Flash, IA32, IA64 and several generations of each.
> AMD won't be able to keep up.

They don't have to do all that much, you know: Hammer is a good enough
product already that they can ride the process-improvement curve just like
Itanic will be doing for the next couple of years at least.

> They haven't the resources.

IBM does, and has plenty of incentive to help ensure that AMD thrives at
Intel's (and particularly Itanic's) expense. If IBM has great (larger)
Opteron systems to offer and no one else does, so much the better: they
won't compete at all directly with IBM's proprietary offerings, but will
definitely help stunt Itanic's general growth (and Itanic *would* compete
more directly with IBM's other products if it gathered sufficient momentum).

So the combination looks like a win for both companies.

- bill

Relevant Pages

  • Re: Portents of Itanium death
    ... > 2.0 GHz certainly isn't going to do much regarding an increase. ... How much a 40% clock-rate boost will help the Xeon MP ... One was a 2.4 GHz system with 8 GB of RAM, ... > Where is Opteron in that space? ...
    (comp.os.vms)
  • Re: Portents of Itanium death
    ... >> Intel has big speed bumps ahead with Xeon MP. ... Should keep Xeon ... >> competetive with Opteron, ... > In competing with Itanic (assuming Itanic turns into something one might ...
    (comp.os.vms)
  • Re: Portents of Itanium death
    ... >>band won't have Opteron as an option until its performance improves as well. ... compared with a 1.8 GHZ Opteron at 1170 SPECint. ... So a 2.8 GHz Xeon MP might acheive the same single ... future Xeon but cheaper on Int. ...
    (comp.os.vms)
  • Re: Portents of Itanium death
    ... as will Opteron if its clock rate increases on schedule. ... Even if Madison attains the SPECint score of 1250 you recently predicted, ... Xeon should equal that at 3.2 GHz and Opteron should equal it at between 2 ... and 2.2 GHz - both of which are scheduled to occur in a few months at most. ...
    (comp.os.vms)
  • Re: How bad are PCI graphics cards?
    ... One option is to sell the dual Xeon motherboard and buy another, but that is going to be more expensive. ... I'm just a bit concerned that I know the particular CPU intensive application I occasionally run, which might use a week or two of CPU time, runs faster on Opteron. ...
    (alt.comp.hardware.pc-homebuilt)