Processor EM computation benchmark

[goxy]

I built this benchmark based on EMAP5 MOM/FEM code to compare different processors in EM computation.

Structure contain:

FEM Volume:
segments along the x axis: 6
segments along the y axis: 15
segments along the z axis: 1
total inner edges: 229
total tetrahetron elements: 450

MoM Region:
total triangle elements: 444
total boundary edges: 666
total metallic boundary edges: 257

num of current source edges: 1

Memory Required: 11.50 M Bytes


If you have new generation processor results upload here!


Format:

****************************
OS/PROC/MEM

results from bench.log
*****************************

OS-Operating System
PROC-Processor, for AMD Athlon XP put real clock in brackets
MEM-Memory, only type and speed are necessary, not quantity

Some my results:

WinME/Intel Celeron 366MHz/SDRAM66MHz

computing Bdd, C, D matrices
time used: 182 sec
computing correction term
time used: 243 sec
computing A matrix
time used: 1 sec
solving the hybrid matrix equation
time used: 204 sec
computing surface currents:
time used: 0 sec

total time used: 630 sec


WinXP/AMD Athlon 2000+(1.67GHz)/DDR266MHz

computing Bdd, C, D matrices
time used: 36 sec
computing correction term
time used: 49 sec
computing A matrix
time used: 0 sec
solving the hybrid matrix equation
time used: 40 sec
computing surface currents:
time used: 0 sec

total time used: 125 sec

 

[Element7k]

Hi Goxy,

here are my inputs on P4 machines. 3 GHZ and 2.4 GHZ both using DDR 333 MHz Ram AND winXP Pro.

Hope it is helpful to you and that the exe file hasn't done any "damage" or placed spies on my systems

Cheers,
ELement7k

 

[eirp]

Very good idea, goxy, Thanks!!

Here is mine - P4-2.66GHz, Memory bandwith 3197Mb/s (I think more than 333MHz), WinXP pro

Time Usage Report:

computing Bdd, C, D matrices
time used: 41 sec

computing correction term
time used: 52 sec

computing A matrix
time used: 0 sec

solving the hybrid matrix equation
time used: 42 sec

computing surface currents:
time used: 0 sec

total time used: 135 sec

Your Athlon is faster!!!

I will try my home computer with Athlon soon.

Eirp

 

[Element7k]

Hummm.. does this mean that the new generation processor with hyper threading technology does help a little or is the software not programmed to use this? My 2.41 Ghz processor does not have this feature.

Cheers,
Element7k

Very good idea, goxy, Thanks!!

Here is mine - P4-2.66GHz, Memory bandwith 3197Mb/s (I think more than 333MHz), WinXP pro
total time used: 135 sec

Your Athlon is faster!!!

I will try my home computer with Athlon soon.

Eirp

 

[eirp]

Just to add, my P-IV support hyper threading. Goxy's Athlon knocked us down!!

I'm really curious about my home Athlon ...

Eirp

 

[eirp]

Here is my another report:

My home Athlon XP 2800+ Barton @2.1GHz, DDR 333MHz, WinXP pro.
Mainboard: Asus A7V333

Time Usage Report:

computing Bdd, C, D matrices
time used: 30 sec

computing correction term
time used: 41 sec

computing A matrix
time used: 0 sec

solving the hybrid matrix equation
time used: 33 sec

computing surface currents:
time used: 0 sec

total time used: 104 sec

Seems that Athlon really wins!!

People, post your benchamark here too!!!!

 

[goxy]

Hi all!

Realy interesting results!

Memory bandwith is on P4 side but that's important for web servers and databases.
Arithmetic Logic Unit is on Athlon side and that's crucial for us.

Best ratio price perfomance has Athlon 2600+(2133MHz)/133FSB.


Emap5 code is compaled from ANSI C, special P4 instruction aren't used and maybe so different.
This "special instruction" was implemented in Feko 4.01 and they should be to improve perfomance for 25% on P4, but on Athlon they decrese for 25%, and because Feko 3.05 is 25% fastest than 4.01 on my Athlon.


Next benchmark will be on comercial software-CST MWS to check this!

Test model is rod antenna from CST examples but overmeshed for better comparation.

Transient solver time results are at the end of log file(Results>View Logfiles>Solver Logfile).
Make calculation only with posted files (wihout previously meshed results), if you want to try again than delete all results and unpack posted files again.


My results:

Athlon 2000+(1.67Ghz)/DDR266/MSI KT4V(KT400)/WinXP

Number of processors used: 1
Mesh generation time: 6 s
Solver time: 79 s
------------
Total time: 85 s (= 0 h, 1 m, 25 s )

 

[eirp]

P4-2.66GHz, Memory bandwith 3197Mb/s (I think more than 333MHz)
MB: ASUS P4G8X, WinXP pro

Number of processors used: 1
Mesh generation time: 5 s
Solver time: 61 s
------------
Total time: 66 s (= 0 h, 1 m, 6 s )

Now Pentium wins

 

[eirp]

My home Athlon XP 2800+ Barton @2.1GHz, DDR 333MHz, WinXP pro.
Mainboard: Asus A7V333

For now the best

Number of processors used: 1
Mesh generation time: 4 s
Solver time: 58 s
------------
Total time: 62 s (= 0 h, 1 m, 2 s )

Regards,

Eirp

 

[goxy]

Thanks Eirp!!

Athlon is our friend :spoko:

Element7k can you post your 3 GHZ and 2.4 GHZ P4 results, we waiting for you for final scores.

 

[Element7k]

P4 3 GHZ DDR 333 MHz Ram AND WinXP Pro.

Solver Statistics:

Number of processors used: 1
Mesh generation time: 5 s
Solver time: 51 s
------------
Total time: 56 s

P4 at 3Ghz only slightly faster then eirp's Athlon at +2800.
However, I noticed that Athlon experience temperature problems during the hot summer if you don't have air-condition. 8O But P4 runs quiet with no temperature problems.

P4 2.4GHz a bit later as it is occupied with computation at present.

Cheers,
Element7k

 

[eirp]

I would like to note that new Athlon's core - Barton is much cooler than previous ones.
Congratulations, Element7k
I must overclock my 2800+ to beat PIV!!

Regards,
Eirp