Re: Sun to use AMD Opteron - announcement expected Monday

From: Rob Young (young_r_at_encompasserve.org)
Date: 11/26/03 

Date: 25 Nov 2003 21:23:23 -0600

In article <O-CdncZiT_qCUF6i4p2dnA@metrocast.net>, "Bill Todd" <billtodd@metrocast.net> writes:
>
> "Rob Young" <young_r@encompasserve.org> wrote in message
> news:ExiFv198tRzB@eisner.encompasserve.org...
>> In article <09SdnV5YyfuYMl6i4p2dnA@metrocast.net>, "Bill Todd"
> <billtodd@metrocast.net> writes:
>> >
>>
>> >
>> > Itanic performance is lack-lustre only by comparison with the
>> > historically-inflated claims for it (and with more recent claims, if you
>> > compare real-world performance with some of the benchmark scores that
> are
>> > being touted for it). In absolute terms, its performance is certainly
>> > competitive - but that's not nearly sufficient to get many people to
> adopt a
>> > new platform in preference to their tried-and-true existing ones.
>> >
>>

>> Why don't you say Itanium is mostly "leading" instead of competitive?
>
> Because it's not leading - just competitive.
>
>> SSL or a Web thing for Opteron? Okay.
>
> That's a good start on explaining why my characterization is suitable.
>
> But Itanium is leading in
>> tpmC (yes - not a single trailing number there).
>
> It trails rather badly behind POWER4+ in per-processor TPC-C performance.

        Per processor tpmC performance? Come on, you're not going to start
        talking about SpecInt per MHz are you? Wouldn't the best comparison
        be total tpmC performance and $/tpmC? After all, that is what
        the TPC council touts. Or realworld, if you have budget, you
        can buy the 4 CPU power or 4 CPU SPARC and associated DB licenses,
        what to choose? Simple.

> We *are* talking about processors, here, right? I mean, the fact that Sun
> (back when it used to submit TPC-C results) took far more processors than
> Alpha to achieve comparable results was something that *you* always chose to
> emphasize...
>
        Still do. Because Oracle is $40000 per CPU and DB2 is $33000
        per CPU. If buying a 4 processor box you can go with the
        under-whelming 4 CPU SPARC or a much higher powered Itanium box.

        So wouldn't the 32 procssor Power4 box make more sense than the
        64 processor SuperDome? Depends. You may want to run a DB that
        delivers the most bang for the buck - MSSQL. At $16541 per
        CPU it helped HP turn in a higher tpmC number than IBM and much
        cheaper, $6.49 $/tpmC for HP, $8.55 $/tpmC for IBM.

        There are several directions we can explore here regarding cost,
        performance , DBs, etc.

>> SAP - yes.
>
> Er, no.
>
> If you've abandoned your historical position noted above that the number of
> processors is significant, then you should note that the top 5 SAP SD 2-tier
> positions are held by Sun and Fujitsu, with high-processor-count SPARC
> systems. If not, let's compare systems with equal numbers of processors:
>
> 6th place is held by (gasp!) a 32-processor Marvel system - though there's a
> good chance that a 32-processor POWER4+ system might edge it out for that
> position (a previous-generation 1.3 GHz 32-processor POWER4 system is only
> two positions behind, in 8th - after another high-processor-count Fujitsu
> system).

        Well - yeah. I wouldn't claim they had the highest SAP numbers
        across the board. Sorry for not being clear there. They do lead
        in 4-way and 8-way. In fact, 2 weeks ago IBM posted a 4-way
        Itanium SAP 2-tier that is 199000 dialog steps/hour. By the way,
        that 32-processor Marvel is in 7th place now, a 32-way NEC passed
        it. That ideasinternational list is dated.

>
> The highest-placing 32-processor Itanic system is a brand-new,
> top-of-the-line 1.5 GHz processor system from NEC - in 10th place, just
> after yet another high-processor-count Sun system.
>

        No. 6th place now. That ideas list is stale.

>> SPECfp2000 - Opteron isn't even close to Itanium.
>
> The one area where Itanic has a clear lead is in raw SPECfp and SPECfp_rate
> performance - though even there one should note that its power consumption
> is high enough that installations with limited cooling facilities can
> achieve equivalent performance per Watt with AMD64 and noticeably *better*
> performance per Watt with POWER4+. Of course, SPECfp isn't really all that
> significant for most commercially-important applications, but I'm happy to
> give credit where it's due.
>

        performance per watt?

        Another area it is a clear leader is 4-way tpmC.

> SPECint2000? AMD
>> has a 1376 base entry, Itanium has a 1322 entry. Opteron squeeks
>> past Itanium there.
>
> Indeed it does - though if you want an AMD *server* processor that beats
> Itanic (which seems like a fairer comparison) you'll have to wait a few
> weeks until the x48 Opterons appear.
>

okay

>>
>> But to characterize Itanium performance as lack-lustre is a stretch.
>
> I guess in your haste to tout Itanic you must have missed the fact that I
> said exactly the same thing (noting that lack-lustre was an appropriate
> characterization only when comparing Itanic performance with with inflated
> historical claims or when trying to relate its benchmark results to
> real-world performance).
>

        Itanium leads in many metrics. To characterize it
        as lackluster is a stretch. SPARC is lackluster, MIPS is lackluster.
        Certainly not Itanium.
        
                                Rob

Relevant Pages

  • Re: Itanium Madison blasts Sun, IBM in encryption specs!
    ... similar capacity to a Sun V280 but well under half the price. ... match a two-CPU non-SPARC system at SPECweb99_SSL) and compare them to ... Base price per Sun's online pricing pages for the V240 appears to be ... Opteron would be $14380, IIRC, you have asserted previously (opening ...
    (comp.os.vms)
  • Re: Booting from a fedora core live cd
    ... and compare with checksums run by the FC guys. ... okay. ... This matches the checksum published by Rahul Patel on 5th Jan 2007, ... Mark Hobley ...
    (comp.os.linux.misc)
  • Re: OT for fun
    ... > | Okay, to follow up -- there were other choices that supposedly match ... > | Andrews, Willem de Kooning, and Helmut Kohl. ... > Also in pic are John P and Sean Chercover. ... I was just grateful it didn't compare me to Jack Elam. ...
    (rec.arts.mystery)
  • Re: Why couldnt you "hi-fi-ize" an old recording with computer technology?
    ... you can compare them and say - okay, overall, what's different? ... the old gear consistently do differently than the modern gear. ... sound like it was recorded in hi-fi, but still retain the character of the ...
    (rec.audio.pro)
  • Re: Retrieve Index of Activeworkbook?
    ... >The only option you have is to use a loop to go through the index numbers ... and compare the name property to the ActiveWorkbook Name. ... Okay, thanks, Leith. ...
    (microsoft.public.excel.programming)