IT-avdelinga 
Startsiden

Til Service/tjen.
 
Hjem
Informasjon
Adresser
Ansatte
Nyheter
Prosjekter
Organisering
Retningslinjer
Tjenester
Brukerstøtte
IT for ansatte
IT for studenter
IT-sikkerhet
Nettjenester
High Performance Computing (HPC)
About us
Projects
User Support
Advanced Support
Application forms
Itanium Cluster
News
Rocks-PBS-Roll
Diverse
 
 
Language:

 
[ Login ]
set width of left cell in netscape 4
Startpage › tjenester › itavd › HPC › HPC-Diverse › 6
Printer friendly


3

Benchmark

Benchmark

This is the UiT benchmarks that is needed for delivery of Tender to UiT HPC cluster procurement in 2007.

Two codes are used for benchmarking: Dalton (version 2.0) and ADF 2006.01.

One test is performed with ADF and two tests with Dalton.
All benchmarks are single cpu runs that test the machine's ability to run several copies of the same program concurrently.

Any compiler can be used, but the system setup with respect to hardware, software and compiler switches used must be submitted together with the results. The source codes should normally not be modified.


Compilation examples and test run examples are done using ifort version 9.0 and icc version 9.0 under Rocks v4.1.0 (aka RH EL 4).

Questions about the benchmarks can be directed to hpc-staff@cc.uit.no.

1.- Dalton


The Dalton code can be downloaded at
https://heimdal.uit.no/hpc/dalton-2.0.tgz
This code can only be used for the purpose of this benchmarking.
Username and password can be requested by contacting us, hpc-staff@cc.uit.no.

Example configs on Itanium2: Makefile.config Itanium2

Note: On itanium the routine "cc_4hyp.F" is compiled (manually) with option -O1 instead of -O3, because otherwise the compiler "hangs". This is not expected to have any significance for the benchmarking results.

See also the dalton homepage for information on the dalton program.

2.- ADF


The adf code can be retrieved for no cost from the download section of the SCM site. To receive a download password and temporary license for benchmarking send a request to support@scm.com.
Vendors should mention in the comment field that they need the source code for the benchmark in Norway.

3.- Setup for benchmarking


All input files used for the benchmarking can be found here.

The tests are performed using the XXX.sh scripts.
The script runs succesively bunches of 4 or 8 copies in parallel, 40 copies in total.
You will also need different input files.
The XXX.txt files are used by the XXX.sh scripts to run the code.
test_Fe.inp and t21* are input data for the adf test.
twoelint_test.dal, 6-31G* and ano-4 are input data for the first dalton test.
numint_test.dal, 6-31G and ano-4 are input data for the second dalton test.

The path in the *.txt files have to be adapted.
The number of concurrent runs used in the *.sh files (CONCRUNS) should be set equal to the number of cores on the node.


The time for a test is obtained by running the script XXX.sh with "time". For example:
time ./Bench_dalton1.sh
The "real" time is then used.
The script gives the time for doing the same calculation 40 times on one node.

Example of correct output for a single run (test1): twoelint_test.out . Check in particular that
"DIIS converged in 14 iterations !" and
"Final HF energy: -387.960627961522"

Example of correct output for a single run numint_test.out. Check in particular that
"DIIS converged in 14 iterations !"
"Final DFT energy: -390.301183526601"
(the energy results may differ in the last digit)

Example of correct output for a single run (adf):
Fe_test.out


4.- Normalized node performance


The normalized time for the test is obtained as (times in minutes for the 40 runs):
tdal = (Time_dalton_test1/44.5 + Time_dalton_test2/37.2)/2
tadf = Time_adf/372

The normalized node throughput is taken as the inverse of a weighted normalized time.
UiT benchmark normalized througput:
Tuit = 1/(0.7*tdal + 0.3*tadf)

A higher normalized throughput corresponds to better performance.


4.1- Examples


4.1.1- Node with 2 Itanium2 Montecito dual, 1.6GHz

40 copies, using all 4 cores.
Time (real) Tdal1: 123 minutes
Time (real) Tdal2: 76.1 minutes
Time (real) Tadf: 685 minutes

Normalized times:
tdal = (123/44.5+76.1/37.2)/2 = 2.40
tadf = 685/372 = 1.84

UiT benchmark normalized througput
Tuit = 1/(0.7*2.40 + 0.3*1.84) = 0.45


4.1.2- Node with 2 Xeon Quad, 2.3GHz

40 copies, using all 8 cores.
Time (real) dalton test 1: 44.5 minutes
Time (real) dalton test 2: 37.2 minutes
Time (real) ADF: 372 minutes

tdal = (44.5/44.5+37.2/37.2)/2 = 1.0
tadf = 372/372 = 1.0

UiT benchmark normalized througput
Tuit = 1/(0.7*1.0 + 0.3*1.0) = 1.0


4.1.3- Node with 2 Opteron dual core, 2.6GHz

40 copies, using all 4 cores.
Time (real) dalton test 1: 86 minutes
Time (real) dalton test 2: 82.2 minutes
Time (real) ADF: 1033 minutes
tdal = (86/44.5+82.2/37.2)/2 = 2.07
tadf = 1033/372 = 2.78

UiT benchmark normalized througput
Tuit = 1/(0.7*2.07 + 0.3*2.78) = 0.44

Would you like to receive e-mails about updates in this folder?
E-post adresse:

Universitetet i Tromsø, 9037 Tromsø, tlf. 77 64 40 00
Updated by Peter Wind on 22.06.2007 at 13:25
Ansvarlig redaktør: IT-direktør Magnar Antonsen


Read this page in: Bokmål

set width of main cell in netscape 4

All UiT-pages
Advanced search

Pages:
Showing 1-3 of 3
investering 2007
Bakgrunn
Benchmark
set width of right cell in netscape 4