Re: 500.000 AMD64's shipped...

From: Dennis (dennis_at_etinc.com)
Date: 01/15/04 

Date: 15 Jan 2004 07:48:02 -0800

I'll take a 3.06Ghz Xeon over either. The stability and performance of
the peripheral chipsets, particularly the bus, are more important, and
intel has a wide lead in that category. All of the AMD chipsets have
flaws. Additionally, most high end machines are I/O bound, not cpu
bound. So for the vast majority of applications your little arguments
and benchmarks are just plain meaningless.

Bravo to AMD however, for doing good marketing. Thank goodness there
is always a benchmark to make your case, no matter what it might be.

DB

Andrew Harrison SUNUK Consultancy <Andrew_No.Harrison_No@nospamn.sun.com> wrote in message news:<bu3gu1$mm2$1@new-usenet.uk.sun.com>...
> Keith Parris wrote:
> > JF Mezei <jfmezei.spamnot@istop.com> wrote in message news:<4003A02E.23A91A6F@istop.com>...
> >
> >>the inferior iA64 thing.
> >
> >
> > Do you realize the AMD x86-64 architecture isn't even a true 64 bits?
> > Yes, it has 64-bit registers, but its virtual address space is only 48
> > bits. That is actually smaller than Madison's _physical_ address space
> > of 50 bits. (Opteron has a 40-bit physical address space.) IA64 has a
> > full 64-bit virtual address space. While x86-64 contains 64-bit
> > extensions to x86-32, Itanium was designed for 64 bits from the ground
> > up.
>
> Does any of this make any difference at all.
>
> 40 bit addressing will allow the Opteron to address more
> physical memory than is currently available in any microprocessor
> based system.
>
> The fact that Madison has 50 bit addressing and Opteron has
> 40 will make absolutely no difference at all to current
> systems.
>
> You need to find a feature that matters Keith 40 bit vs 50
> bit physical addressing doesn't.
>
> You have however found a feature that does matter, Opteron is
> based on x86 Itanium isn't, Opteron has all of the current
> IA32 SW catalogue available to it Itanium has ~500 apps
> for each OS it supports. Thats a feature that does matter
> but not in the way you hoped for.
>
> >
> > Integer registers: 16 for x86-64; 128 for Itanium.
> >
>
> But Opteron Integer performance is stubbornly faster than
> Itanium despite being saddled with 1/8 the registers.
>
> 1477 SPECint vs 1322
>
> So you have found another feature that doesn't matter.
>
> You should also have know that the measure was wortheless
> because Itanium needs more registers anyway.
>
> > Floating-point registers 8 for x86-64 (it's saddled with the
> > now-ancient x87 floating-point architecture); 128 for Itanium.
> >
>
> Ahh finally you have found a feature that might matter though
> it again may not be the number of registers.
>
> The trouble is that this one feature thats positive for
> Itanium doesn't actually matter to the vast majority
> of people who will buy your systems.
>
> As ever a misplaced effort on your part has been rewarded
> with a poke in the eye.
>
> regards
> Andrew Harrison
> > If you look at the SPECfp2000 results at http://spec.org, all the top
> > results, from HP, SGI, Bull, Dell, Supermicro, and ION Computer
> > Systems, are with the Itanium 2. The #1 top result is for an HP
> > system with Itanium 2 (1.5 Ghz) at 2119. Power4+ (1.7 Ghz) is way
> > behind at 1699, Pentium 4EE (3.2 Ghz) at 1516, Opteron (2.2 Ghz) at
> > 1514, EV7 (1.15 Ghz in GS1280) at 1482, and SPARC (1.32 Ghz) at 1350.

Relevant Pages

  • Re: 500.000 AMD64s shipped...
    ... The fact that Madison has 50 bit addressing and Opteron has ... You have however found a feature that does matter, ... based on x86 Itanium isn't, Opteron has all of the current ...
    (comp.os.vms)
  • Re: MacBook Air
    ... Dimensions really matter a great deal ... Not something about function; just a design choice. ... sacrificing a feature or behavior. ... and it's a flip phone. ...
    (comp.sys.mac.advocacy)
  • Re: swap-prefetch: A smart way to make good use of idle resources (was: updatedb)
    ... It doesn't matter for who the feature is intended, ... Perhaps one of the reasons is that this is core kernel code. ... And there don't seem to be many people opposing swap prefetch either. ...
    (Linux-Kernel)
  • Re: Difference between Feature , Function & Module
    ... Responding to PopCorn... ... Feature = A feature is something which make ur application stand ... Those rules and policies may or may not map 1:1 to software modules and procedures. ... A better way to describe it would be that a module captures a cohesive subject matter. ...
    (comp.software.testing)
  • Re: Dumb find Question
    ... merely pissing in the same pot;-) ... As a matter of fact I have - you just talked about it;-) ... > about implementing a feature. ... The problem with the world is stupidity. ...
    (alt.os.linux)