From: Holger.VanKoll@SWISSCOM.COM
Date: Mon Nov 04 2002 - 12:48:15 EST
> Clock speed is of very limited use for comparing processor power,
unless the chip architecture is very similar: more similar than Power4
vs Power3.
Sure, no doubt.
AFAIK intel reduced the number of pipelines in the pentium4 just to be
able to run it at a higher clockspeed. At this higher clockspeed its
slower than the old version, but marketing dominates at some companies;
especially those targeting at end-customers (not at companies).
I would not expect IBM to do something similar, however.
> I would also be very wary of your own tests: there are a lot more
factors involved in performance than the CPU.
Its doing big sorts in memory. Only cpu/memory counts here. There is no
paging.
Same for my gzip-tests.
> For a DB application, I/O speed is probably more important than CPU
power.
Yes, but is there anything io-speed related (fc-adapters, disks etc.)
that I can buy for a p4 but not for a p3?
If there is any io bottleneck, cpu-speed is of course not important
anymore. But (here) there isnt.
> You'd do better to find a benchmark for a load similar to that which
you will be running, if you want to compare them.
Thats a big part of the problem. All IBM gives to you is rperf. I must
size my systems given on one benchmark-number.
You can get some more numbers (spec.org etc.) that all have at least
some use for sizing.
>Is your actual application running slower than expected?
No. Depending on several parameters it runs a bit faster/slower; no
problem here.
I just wonder what the modifications / news in the p4 are for.
Compared to RS64-IV or -III you have fewer l1-cache (to be able to clock
higher?) f.e. But you have a big l3-cache on p4 and only there.
The p4 is the fastest cpu you can get from ibm. It needs much more
clock, but thats not important.
Anyway, if you offer systems (especially when upgrading from older
ibm-systems) non-technician decisionmakers see mhz and ask questions.
In a perfect world they wouldnt - meanwhile it would be nice if I could
give a short explanation of this high-clockspeed needs.
Simon Green
Philip Morris ITSC Europe
AIX-L Archive at http://marc.theaimsgroup.com/?l=aix-l&r=1&w=2
AIX FAQ at http://www.faqs.org/faqs/aix-faq/
N.B. Unsolicited email from vendors will seldom be appreciated.
-----Original Message-----
From: Holger.VanKoll@SWISSCOM.COM
[mailto:Holger.VanKoll@SWISSCOM.COM]
Sent: 04 November 2002 16:17
To: aix-l@Princeton.EDU
Subject: power4 performance
Hello,
does anybody have some real-workload performance data of
p670/690 or whatever runs a power4 that he likes to share?
I am a bit disappointed about the performance of my p670
@ 1100 mhz... its only double as fast as a sp 9076-power3-II @ 375mhz.
So its half as fast per mhz.
I measured it using self-written applications and
aix-tools like gzip... same results.
If you look at
http://www-1.ibm.com/servers/nl/eserver/pseries/p690/p690.html
<http://www-1.ibm.com/servers/nl/eserver/pseries/p690/p690.html>
you see a rperf of 50.56 for a 32-way-1100mhz-power4 and
a rperf of 27.65 for a 24-way-450mhz-rs64IV.
Double as much for 32x1100mhz as for 24x450mhz.
That means 32 power4 @ 1100mhz are as fast as 48 rs64iv
@ 450mhz. Or 35200 power4-mhz equal to 21600 rs64iv-mhz.
Of course these are only benchmark-numbers. Thats why I
am asking for real-life/workload experience.
I would not expect a next-generation cpu like power4 to
need 1.5 or 2 times the clockrate of the older model to perform the
same.
This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 22:16:18 EDT