The performance of HFSS V11.

EDAboard.com Forum Index -> Electromagnetic Design and Simulation -> The performance of HFSS V11.

Author

Message

cr_skyworks

Joined: 04 Aug 2006
Posts: 33

13 Jun 2007 22:35 hfss 11 manual

Anyone have done some computations using HFSS V11? Just curious about the real performance.

Azulykit

Joined: 18 Feb 2007
Posts: 851
Helped: 103
Location: All over the Map.

14 Jun 2007 21:27 hfss 11

I have been fooling with a 11.0 beta version and am pleased with the results. The beta crashes once in a while, but that is par for the course and expected.

Real work keeps getting in the way, so I do not have more than a general impression.
Their previous releases have all been improvements from the previous revisions and I expect to see the same here (no, I do not work for Ansoft).

dingjingfeng

Joined: 12 Aug 2005
Posts: 196
Helped: 9

15 Jun 2007 10:15 hfss v11 tutorial

I am waiting to secret of HFSS V11
ha!

cooleyn

Joined: 22 May 2004
Posts: 40
Helped: 5

15 Jun 2007 13:26 hfss 11

Glad to see the improvement in HFSS 11. Looking forward to using it in future.

Azulykit

Joined: 18 Feb 2007
Posts: 851
Helped: 103
Location: All over the Map.

17 Jun 2007 18:53 hfss ram

I had some free time and ran this comparison on a filter problem. The same setup and PC (64 bit Xeon, MS XP Pro 64 bit)

Version 10.1.2: 56K tetrahedra, 1.4 GB memory usage, 20 minutes CPU time
Version 11 beta: 43K tetrahedra, 1.2 GB memory usage, 14 minutes CPU time

I am a little curious about why the number of tets decreased. I did not change the setup between the two runs. Both ran 16 adaptive passes and then performed a fast sweep.

The task manager reported a 1.7 GB RAM usage during the beta run. I did not watch the 10.1.2 run as that model I ran last week. I usually see a greater usage reported by the task manager than in the HFSS profile. I suspect that is because operating system overhead is probably included.

I wouldn't take this to be the final word as the beta is still a work in progress. I was pleased by the comparison. It looks like Ansoft has done a good job.

wave-maniac

Joined: 05 May 2001
Posts: 279
Helped: 3
Location: USA

17 Jun 2007 19:03 hfss v11

Dear Azulykit

Thanks for the info. I guess that HFSS V11 beta might have used the higher order basis functions as default setting, hence the reduction of number of tets. I would be grateful if you provided some benchmarks about iterative solver, i.e. direct solver vs. iterative solver, particularly for high number of tets, e.g. 200k.

Regards

rfmw

Joined: 10 May 2001
Posts: 520
Helped: 38

18 Jun 2007 0:20 hfss tutorial v11

Yeah, thanks Azulykit for the test-bench results. Still, you must also provide us with the info about your RAM in the machine. And if both versions got to the same delta.

However, where the hell is the famous sentance: "V11 uses half the memory of the previous versions for the same structure". If only 200MB of 1400MB memory requirement reduction, thats not much if you ask me.

Azulykit

Joined: 18 Feb 2007
Posts: 851
Helped: 103
Location: All over the Map.

18 Jun 2007 5:06 hfss v11 tutorials

"Still, you must also provide us with the info about your RAM in the machine."

I must? I ran this particular set of simulations on a platform that has 5 GB of RAM installed. XP Pro 64 bit OS.

"And if both versions got to the same delta. "

The older ver. 10.1.2 software reported a Max Mag Delta S of 0.010701 while the ver. 11 beta reported 0.0097475. I limited the number of adaptive passes to 16 and neither reached the target of 0.005.

"However, where the hell is the famous sentance: "V11 uses half the memory of the previous versions for the same structure".

Who said that? Cutting the RAM requirement by -3dB would be something. Or should it be -6 dB? Does anyone know if RAM is power or voltage? Just kidding.

The pair of runs were relatively easy to make and my intent was to add a bit of specific information to my earlier comments. I am in the middle of a time/work crunch so I unfortunately do not have the time to do extensive benchmarking. The reality of this is that when the latest v11 is released I will migrate there. I am not sure that extensive before and after comparisons are really worth the effort. I plan on spending a big part of my time studying the features of the new software. I like what I have seen and as usual the interfaces have changed a bit and are causing some consternation.

I understand that Ansoft intended to release v11 at MTT. I think they are doing the final polishing and we will see it soon. Has anyone seen it yet?

I hope this is helpful.

rfmw

Joined: 10 May 2001
Posts: 520
Helped: 38

18 Jun 2007 11:20 hfss v11 tut

"I must? I ran this particular set of simulations on a platform that has 5 GB of RAM installed. XP Pro 64 bit OS."

Well, if your PC had only 2GB of RAM, that would explain why the difference between computation time, wouldn't it?

And about the half-the-RAM usage, that is written in the Ansoft's HFSS V11 release news...

sassyboy

Joined: 20 Sep 2003
Posts: 88
Helped: 14
Location: United Kingdom

18 Jun 2007 13:28 hfss v11 differential waveport

I understand the result is completed on a 64bit OS. I am still 32bit user. Do HFSS 10 and 11 guarantee to run faster if the same 32bit OS computer is replaced with a 64bit OS? What about dual core capabilty? Does HFSS 10 and 11 automatically detects this and make use of this? Or is this an extra option for the standard license price? I am thinking of getting a quad core PC for the company, does this mean I have to purchase 4 licenses to accommodate 4x processing power.

Thank you in advance.

Google
AdSense
Google Adsense

18 Jun 2007 13:28 Ads

cr_skyworks

Joined: 04 Aug 2006
Posts: 33

18 Jun 2007 20:08 scale3dsettings, hfss 11

If the HFSS V11 is not much better than HFSS V10, some customers won't spend the extra money to update.

rfmw

Joined: 10 May 2001
Posts: 520
Helped: 38

19 Jun 2007 22:37 hfss v11 directlinks

I'm very disappointed in HFSS 11beta. Simulating the same project file with v10.1 and v11beta, the v10.1 made all passes, while v11beta stopped with insufficient memory error at the final pass. The project wasn't changed in any way. The structure is electrically large, with approx. 70.000 tetrahedra in the final pass. So the Azulykit's benchmark is OK after all! V11beta doesn't bring much difference!!!

Ok, I said that maybe iterative solver within v11b will reduce memory appetite, but unfortunately the result is the same. Perhaps iterative solver can't help with my structure, but the feeling is bad anyway, since v11beta doesn't bring help for me...

Azulykit

Joined: 18 Feb 2007
Posts: 851
Helped: 103
Location: All over the Map.

20 Jun 2007 1:04 hfss price

On the plus side I saw a reduction of solve time from 20 minutes to 14 minutes and a reduction in memory (1.4 GB to 1.2 GB) usage.

That is helpful.

I want to see how the higher order basis functions work out as well as the iterative vs. direct solver.

What other features have been added?

I am hopeful but holding my judgement until I have more experience with this latest version.

I just noticed the quad processor license question. My understanding is that one needs to purchase a multiprocessor license to gain an advantage when using a platform with multiple processors. I would speak to Ansoft on this point as I am not a spokesman for them. I admit that I find their code very useful.

aw

Joined: 09 Jun 2001
Posts: 58
Helped: 2

30 Jun 2007 13:37 optimetrics hfss v11 tutorial

Hi,

although any new features are to be welcomed I just don't see what the improvement is here. Are the comparisons that you have made based on the new higher order tets and the direct solvers or not? This would be necessary before everybody jumps to conclusions. As for electrically large models, I am not sure that a halving of memory (if anyone has seens this ratio this besides Ansoft themselves) helps for a method which does not scale linearly with mesh cells anyway. For electrially small models, it may help.

aw

rfmw

Joined: 10 May 2001
Posts: 520
Helped: 38

25 Aug 2007 19:08 learn hfss v11

Anybody noticed bugs in the v11? The computation of Zo values for coupled microstrip line (i.e. differential line) feeded by wave port is incorrect. v10 does it correct.