Re: Sparky is losing the race Andrew ... and badly!

From: Andrew Harrison SUNUK Consultancy (Andrew_No.Harrison_No_at_nospamn.sun.com)
Date: 05/16/03 

Date: Fri, 16 May 2003 17:57:15 +0100

Bob Ceculski wrote:
> Sun might as well forget it Andrew ... you should have
> bought alpha/vms/tru64 when you had the chance ...
> NIH (not invented here) syndrome will hurt the company!
>
> http://www.theinquirer.net/?article=9437

Umm IBM has announced 1.7 GHz power 4 systems.

They are shipping them later this year.

Sun announces for immediate shipment.

When you can buy and install a Power 4 processor
running at 1.7 GHz then get back to me, otherwise
its not interesting.

You also cannot buy Itaniums running at 1.5 Ghz
when you can they may also be interesting but
currently they arn't either.

And sadly the Register article as usual betrays
a total lack of understanding about what actually
makes a difference from a performance standpoint.

Take Power 4 1.7 GHz SPECfp performance.

First the single CPU performance is with 7 of
the 8 CPU's in an MCM module turned off so that
the 1 remaining CPU gets access to 128 MB of
cache.

Like to take a guess how much difference this
makes to SPEC performance.

Like to take a guess how many IBM P670 and
P690 customers only use 1 CPU. No don't bother
on that one there are none.

So lets take a look at SPECrate because thats
power 4 without the CPU's turned off to get
more cache.

Humm ohh dear a 32 way Power 4+ machine that you
will be able to buy sometime does 372 SPECratefp
when compared with a 32 way F12K result of 338
which we are currently shipping. Gosh 10% faster
for a system that you can't currently buy. Hardly
compelling. You can buy a 36 way F12K now which is
faster than the P690 will be when it ships.

Or put another way Power 4 sure as hell isn't
anything like as fast a TheRegister and you
think it is when people actually start using
in systems for the job that the system was designed
for. Whatever you and TheRegister may think
IBM never designed the P690 as a single CPU
server and none of their customers use them as such.

Or how about the 1450 MHZ P650 8 way result funny
its under half the performance of a 16 CPU F12K
for SPECratefp. Or put it another way Power 4
sure as hell isn't faster than SPARC on a per
CPU basis, even when comparing a small IBM
with a nice short backplane with a huge Sun
with a dirty great long backplane.

So whats happening.

1. You got fooled by meaningless single CPU benchmark
numbers at did TheRegister (again). When we are talking
about a 64 bit market which is almost all about server
performance with larger SMP systems.

2. You ignored system balance (again). Having a balanced
system design for example low latency high bandwidth
backplanes makes more of a difference than you seem
capable of comprehending.

3. You confused what Sun, HP, IBM etc can ship now with
what some vendors choose to pre-announce in the hope that
it will slow down their competition (again).

Until you can buy it HP etc can't revenue it. You and
TheRegister would look much less ridiculous that you
currently do if you grasped that simple concept.

Thanks for yet more evidence that you and The Register
need to get out a bit more.

regards
Andrew Harrison

Relevant Pages

  • ASH: ASH #98: Rising Sun Part 2 - I Shall Return
    ... The cover shows a uniformed figure with a corncob pipe standing on the ... shore in front of a swollen red setting Sun. ... figure and lit by the red sunset, are the members of the Freedom Alliance. ... He had the raw power to sweep out large swaths of the moth-shell, ...
    (rec.arts.comics.creative)
  • ASH: ASH #100 - Starslayers: Epilogue and Notes
    ... The dimmer second Sun that had replaced the giant orange Sun was ... Like most assortments of orphans and abandoned children in the Lower City, ... fraction of the stolen power, I would have been able to tell. ... because Green Knight was the only one of them not to assault the Fornax. ...
    (rec.arts.comics.creative)
  • Re: a7v600-x stuck in reboot loop?
    ... [Motherboard Specification] ... [CPU Specification] ... So I hold the power button in till computer shuts off ... once or twice sometimes more before the video initializes. ...
    (alt.comp.periphs.mainboard.asus)
  • [RFC PATCH v2 0/7] Tunable sched_mc_power_savings=n
    ... The existing power saving loadbalancer CONFIG_SCHED_MC attempts to run ... to keep rest of the CPU packages idle for longer duration. ... The power savings and performance of the given workload in an under ... * cpu package for power savings ...
    (Linux-Kernel)
  • Re: [PATCH]cpuset: add new API to change cpuset top groups cpus
    ... I have tried few load balancer hacks to evacuate cores but not a solid ... where the word "power" is incorrectly used in the thread above) ... aggregate idle time on cpu/core/packages so as to maximize the idle time ... other major issue for using cpu hotplug for this purpose. ...
    (Linux-Kernel)