Processing power comparison

[Luis Miguel Parra Chica]

Is it possible to compare the processing
power between several families? I mean,
is it possible to affirm: [ for instance
] the performance of a 6CPUs-K380 is the
same that the performance of a
4CPUs-L2000.

Is any table, any web available with
that kind of data? Are that sort of
comparisons reliable?

You can see some information about
that on:
http://www.hp.com/products1/unixservers/midrange/kclass/infolibrary/index.html
http://www.hp.com/products1/unixservers/entrylevel/lclass/specifications/index.h
tml

but is NOT enough. The main point is the
comparison.

Thanks in advance

[Alan Riggs]

http:\\tpc.org

[Andy Zybert]

When modelling workloads on different models/manufacturer/configurations of kit, the specint_rate95 is usually used.

HP/SUN/INTEL etc subscribe to this methodology by submitted specint_rate's on their models at different cpu levels against a set of benchmarks.

I think the site is:

http://www.dl.ac.uk/TCSC/disco/Benchmarks/spec.htm

Currently any modelling is done using specint_rate95 but the newer specint_rate2000 will be used when sufficient results have been posted.l

[Andy Zybert · Accepted solution]

My own records indicate the relative performance of the models mentioned using specint_rate96 comparisions are:

K380 x 6cpu =935
L2000 x 4cpu = 1186

Therefore in processing power alone the L2000 system should be 20% more powerful.

I would suggest you read the individual test specifications at the site mentioned to ensure comparable systems were used i.e memory/disk subsystems/cache etc

[Philip Chan_1]

Hi,

You can compare their SPEC rates, which mean their SPECint_rate95 and SPECfp_rate95 rates.

SPECint_rate95 rates for many different machines (HP, Sun, IBM, Alpha, ...etc.) can be found at

http://www.specbench.org/osg/cpu95/results/rint95.html

For SPECfp_rate95 the link is

http://www.specbench.org/osg/cpu95/results/rfp95.html

Sometimes you may not find the exact hardware looking for, in that case you can pick some reference models. For example, K380/6 way isn't on the list, but you can reference the SPEC rates for K580/6 way instead because their processing throughput are the same. Same thing apply for L2000/4 way, the rates for N4000/4 way can be used.

Eg.
int_rate95 fp_rate95
K580/6 902 604
N4000/4 1209 1495

Do some simple maths on the above figures then you'll know how much faster the N4000 is. Presuming your applications aren't I/O bounded, performance projection by these figures are very accurate.

Rgds,
Philip