Benchmarks
 

Hi all,

There's been a lot of talk on CFD-Online about which processor is faster and what is best system. So let's use CFD-Online to do some benchmarking of different systems.

Below is a simple little command file which should work for CFX 4.1, 4.2 and 4.3. It models the square lid driven cavity at Re=1000. It is a 2D simulation which runs for a few minutes and only uses a few MB RAM. You should be able to take the command file, transfer it to your system and run it to see how fast your system is compared to everybody else's. Please post your results here on CFD-Online for all to see - and make sure you describe your system sufficiently so we know what the result is from. It is only a simple simulation which will not be representative of all problems but it will be at least a guide of relative speed.

The command file will not converge properly using single precision maths on my machine. It does not have the accuracy to converge to the specified critereon.

To start it off, I have run it on the system we have here at the university. Its specifications are:

CFX version: 4.3 double precision Machine: Sun Enterprise 3000 OS: SunOS 5.6 CPUs: 6x 333MHz (but the run was only on 1 processor, not parallel) Memory: 1536MB SpecFP95: 20 (see http://www.spec.org/osg/cpu95/results/cfp95.html)

Command line: runsolve4 -command m01.fc -release 4.3dp -enviro 01 -dir ./.cfx_tmpdir

Results: Time: 180 seconds Iterations required: 1187

Command file: (It looks like the command file has got mangled in the browser, you may need to do some minor edits to get it to work eg replace the :> with >>)

>>cfx4
:>options

two dimensions

rectangular grid

laminar flow

isothermal flow

incompressible flow

steady state
:>model topology
:>create block

block name 'BOX'

block dimensions 50 50
:>create patch

patch name 'WALLLID'

patch type 'WALL'

block name 'BOX'

high j
:>model data
:>differencing scheme

u velocity 'QUICK'

v velocity 'QUICK'
:>physical properties

>>fluid parameters

density 1000.0

viscosity 0.001
:>solver data
:>program control

maximum number of iterations 10000

mass source tolerance 1.0e-7
:>create grid
:>simple grid

block name 'BOX'

dx 50 * 0.02

dy 50 * 0.02
:>model boundary conditions
:>wall boundaries

patch name 'WALLLID'

u velocity 0.001
:>stop

Re: Benchmarks
 

Good idea. Here are my data:
Sofware: CFX-4.3 NT4.0 (Build 1381: Service Pack 6). Hardware: Compaq Deskpro PentiumIII 500MHz with 576 MB RAM. Results: TOTAL CPU TIME = 221. SECONDS, 1187 Iterations Notes: The CFD calculation writes with a high frequency to the harddisk, therefore the HD also has a major influence on the time required for the calculation.

-Michael Bo

Re: Benchmarks
 

Here's two results from 2 different Sun Ultra Sparcs NB, both carried out while doing other filehandling etc.

Software: CFX-4.3, SunOS 5.7 Hardware: Sun Ultra Sparc 2*300MHz (only 1 used) with 1.5GB RAM. Results: TOTAL CPU TIME = 202. SECONDS, 1187 Iterations

Software: CFX-4.3, SunOS 5.7 Hardware: Sun Ultra Sparc 2*400MHz (only 1 used) with 512MB RAM. Results: TOTAL CPU TIME = 178. SECONDS, 1187 Iterations

Anyone got one of the new Suns? The comparison would be very interesting.

Peter

Re: Benchmarks
 

Here is what I obtained on my systems:

---- System 1 --------------------------

PENTIUM II 450 MHz 256 MB RAM Windows NT 4.0 Build 1381 - Service Pack 5

CONVERGENCE IN 1187 ITERATIONS

TOTAL CPU TIME = 305. SECONDS (Dynamic plot: YES)

TOTAL CPU TIME = 232. SECONDS (Dynamic plot: NO)

---- System 2 --------------------------

AMD Athlon 850 MHz 512 MB RAM Windows NT 4.0 Build 1381 - Service Pack 5

CONVERGENCE IN 1187 ITERATIONS

TOTAL CPU TIME = 152. SECONDS (Dynamic plot: YES)

TOTAL CPU TIME = 127. SECONDS (Dynamic plot: NO)

Re: Benchmarks
 

Here are the results from my test

Compaq Ap500 dual cpu PENTIUM II 450 MHz ½G RAM Windows NT 4.0 Build 1381 - Service Pack 5

1187 ITERATIONS CPU TIME = 240. SECONDS

Compaq Presario 5838 (my home pc for game playing)

AMD Athlon 500 MHz 128 MB RAM Windows NT 4.0 Build 1381 - Service Pack 5

1187 ITERATIONS CPU TIME = 159. SECONDS

The memory is not so important for this test case, It could be interesting test a case with more memory usage to see the difference in 100 and 133MHZ ram and BUS speed.

Regards Jan Rusaas

Re: Benchmarks
 

This particular Athlon vs. Pentium benchmark is extremely interesting. I've wondered about a CFD benchmark, rather than FPS scores on Toms hardware site. Looks like the Athlon is winning the race hands down. Die intel die.

Re: Benchmarks
 

Software: CFX-4.3, IRIX 6.5 Hardware: Origin 2000, 8*R10000 250MHz, 8GB RAM Results: TOTAL CPU TIME = 108.SECONDS, 1187 Iterations