微波EDA网,见证研发工程师的成长!
首页 > 研发问答 > 微波和射频技术 > 电磁仿真讨论 > hfss multiprocessor

hfss multiprocessor

时间:03-30 整理:3721RD 点击:
Hi, guys:

My company is gonna buy a dual-core computer for HFSS, and we don't have the multiprocessing license from Ansoft.

My question is how HFSS uses the dual-core to run simulations without that multiprocessing license.

Say I have a project running on HFSS. And one of my colleagues wants to run another project in HFSS. Will HFSS use both cores to run both projects or it will only use one core to handle the job?

Thanks!

i think HFSS will use one core to run one project, and use another core to run another project, if the memory is enough.

If you have a single seat license. only one of you can use the software at a time.

zhlc3

Thank you for your reply!

I can confirm this.
One license -> One core

If you have more licenses, its one core for each project solver. The sweep can be distributed to more cores, but uses one license each core.

one license does equal one core, but they do have seperate liscense to enable as many processors as you want.

i don't know about the sweep being distributed if you have more than one license. I haven't heard of that or tried it.

in the same computer with dual core Technology you must say to hfss that you have more than one processor so you must go on the hfss like these

tools/options/hfss_options

then clik on SOLVER and on number of processors put the number of CORES that you have

ex: if you have one processor with 2 cores put 2
if you have one processor with 4 cores ( quad core ) put 4
if you have TWO processors with dual core Technology than put 4.

it works and speeds up the simulation.

the lisence is needed when you have more than one MACHINE runing it.

now my time ...
do you know how to simulate the axial ratio for a circular polarized antena on hfss?
thx

The post processor has a feature built into the far field reports to calculate axial ratio.

Thx for your answare but .. what variables i must use ?
and the setup for the swep ?

did you have any example ?

thx

Hello

I have a Dual Core CPU and have tried to compare having entered the numbers of processors to 1 and then 2 in the SOLVE menu under HFSS_options.

With 1 processor i get a "CPU time" and "Real time" that are almost equal.

With 2 processor i get a "CPU time" and a "Real time" that are not equal. Now the "CPU time" is increassed by x, but the "Real time" is decreased by the same amount. I do not know if this is just coincidence, but does this mean that the simulation is faster.
Did not measure the time with a stop watch :)
But is this "Real time" the actual simulation time?

Regards

The real time should be the stopwatch time.

Ideally CPU time should remain constant as you increase the number of processors. However, this is never the case, but it does give one an idea of the threading efficiency. In Intel machines all cores share the same bus to memory so I would expect a linear increase in the CPU time with number of cores. I recently converted by MoM code for multiple threads, and this memory bandwidth issue really hurts performance.

Hello Wiley,

This is interesting.
So if the number of cores (in one CPU) incresases, will there be a point where you will not benefit from the increasing numbers of cores, due to lack of memory bandwidth?

Regards

Hi,

Be aware of the 2GB memory limit for windows XP. This is a bigger problem than the number of Cores. You may be able to lift this to 3 GB ( look for 3GB switch) on the Microsoft website.
If you have XP 64 bit or Windows Server 2003, this will not apply.

HFSS does not run on Vista.

:D

Yep, exactly. Memory bandwidth will not be the only issue limiting performance, but it's one that the programmer has very little, if any, control over.

Although I should be more specific on one point I mentioned in my previous post. For Intel machines, each processor has its own path to memory. Every core of that processor will share that path. So a two dual core processors out performs one quad core processor because the former has two paths to memory.

sorry to use this HFSS topic to ask about CST 2006b...is CST capable doing the same of HFFS (dual core= two projects), in a 64bit OS?

thanks.

I have dual quad core XEONs running HFSS11 on Vista64. I did a quick test on a 3 element dipole array simulation using 1 core and 8 cores. The multi-core is apparently only used in matrix solving. The total solution time was 12:56 for 1 core and 8:33 for 8 cores. Here is a snippet of the profiles. You can see the MCS8 solver is 35 sec versus 90 sec for the MCS1. My guess is MCS stands for Multi Core Solver. If you monitor CPU load with the windows task manager, you can see it nailing all 8 cores during the matrix solves of the adaptive passes and fast sweeps. So for your typical projects, if you add up all the time doing matrix solves, you can estimate how much those extra cores will help. HFSS does not require extra licenses for this multi-core solve; I believe that pertains to clustering.

How did you switch between 1 and 8 cores?

There is an option in Tools -> Options -> HFSS Options -> Solver about multicore usage.

"HFSS does not require extra licenses for this multi-core solve" I believe you DO need a license option for multi-CPU.

Hi madengr, I noticed very similar behavior on a dual quad-core Xeon, running HFSS 10 or HFSS 11 beta on WinXP SP2. Running more than two cores didn't improve the solve time very much, but it sure did chew up all the CPU power. The whole computer became horribly sluggish when I enabled all 8 cores, so I stepped it down to 7. I think HFSS is simply broken. This machine has no problems getting smooth nearly-8x performance with other multiprocessor apps.

I too have 2x 4-core Xeons and running HFSS with full 8-core usage is not sluggish at all. Perhaps you've had harddrive swapping?

Copyright © 2017-2020 微波EDA网 版权所有

网站地图

Top