Moldflow Insight Forum
Welcome to Autodesk’s Moldflow Insight Forums. Share your knowledge, ask questions, and explore popular Moldflow Insight topics.
cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

Hardware specifications for achieving more than 9 million mesh elements

7 REPLIES 7
Reply
Message 1 of 8
scd310
1060 Views, 7 Replies

Hardware specifications for achieving more than 9 million mesh elements

We need to achieve a maximum mesh density using Moldflow Synergy 2012 for better accuracy and analysis. My current system specificaiton are HPZ400, 6GB RAM, 2.67 GHz processor, Quadro 600 (1GB DDR3) Graphic card.

 

While doing meshing over 6 million mesh elements, i get a warning and error message in my logs that GPU memory is not enough. 

 

** WARNING 302151 ** The GPU card memory is not large enough to run the full
                     model.  Therefore, the GPU card will only be used for some
                     time steps when the model is partially filled.


It takes more than 30Hours (meshing + analysis) to complete.Most of the times meshing fails at ~95%,98%..


I am thinking of expanding the RAM to 16GB and upgrade the graphic card to FX1800. Will that help getting better accuracy and results for high density mesh (~9 million elements)? Can some one suggest me the system requirements for getting mesh with over 9 million elements and also to improve speed of process?

 

Should we download some additional patch for upgrading the software?

7 REPLIES 7
Message 2 of 8
yannick.moret
in reply to: scd310

Hi,

You didn't mentionned the processor of your machine and the number of cores you used for the analysis.

 

Adding memory will be useful only if you can see that your machine is swapping (CPU are not fully used by flow3D process because computer is using disk rather than RAM as there is not enough RAM available).

 

My suggestions would be :

- Ensure you run the analysis with multithreading.

- Use the job manager batch queue to launch the meshing and the analysis. So prepare the job, then close your model (that will free up some memory) and once you see the amount of memory allocated to synergy went down(you can even close synergy and reopen it), launch the job in batch queue. Don't reopen the model.

This way you will be sure you have a maximum amount available for analysis.

 

Key is that if the meshing or analysis doesn't have enough space in RAM, it will be slow down dramatically.

 

Adding memory will help you anyway to be more confortable, so you can run bigger size of model and doing pre/post porcessing at same time. If you see the PC swapping (constant disk activities) for your 6 millions model, you definitely should add RAM.

Changing card can help you with post processing, but I thought Quadro 600 would be better as more recent than FX1800 ???

 

GPU is not used in meshing, only for analysis, and you will need a big card to run a 9 millions model...But as the message states, we use it as long as we can. Closing the model in synergy will free up more memory on the GPU too.

You will speed up more analysis if you use multi-threading.

 

Regards

 

 

Message 3 of 8
scd310
in reply to: yannick.moret

Thank you so much for your response.

 

I think your reply will definitely help us, we have already ordered new 24Gb RAM and we will perform some testing before buying new GPU.

 

The current processor we are using is--

 

Intel® Xeon® Quad-Core Processor W3520 (2.66 GHz, 8 MB cache, 1066 MHz memory)

(4 cores)

 

Also is Moldflow Insight 2011 muti thread supported? If so how can we check it/activate it?

 

Regards

Samarth Desai

Message 4 of 8
nordhb
in reply to: scd310

Hi,
yes, for AMI2011 and AMI2012 the 3D Fill+Pack process setting are default to use:
- GPU if available,
- and number of threads for parallelization set to Automatic
(Automatic parallelization enables the solver to read the CPU usage at each time step and assign an
appropriate number of threads to use at that time step of the analysis. Therefore, the number of threads
used in the analysis may vary if the machine load (eg: CPU usage) varies during the analysis.)


You find this in Process settings wizard such as:
Process settings wizard Fill+ Pack : Advanced Options... Solver parameters;Edit / tab Fill+Pack Analysis / Solver parameters...
And if Warp sequence is added:
tab Warp / number of threads for parallelization

It will also be seen in screen output, in Matrix Solver Setting.

And when running analysis, you can see cpu load in Task Manager.
Topic http://forums.autodesk.com/t5/Autodesk-Moldflow-Insight/Flow-analysis-using-only-8-of-CPU/td-p/32202...
will give some more information.

Regards,
Berndt



Berndt Nordh
Message 5 of 8
fukushige
in reply to: scd310

Hi, all

 

I've tried running 3D fill analysis with 1.7 and 10 million mesh model using the following hardware.

 - Windows 7 Pro, 64-bit

 - Intel Xeon CPU X5690 3.47GHz (2 processors)

 - 24GB RAM

 - 2GPUs (Tesla C2075 & Quadro 2000 are installed)

 

When I ran 1.7 million mesh model, AMI2012 used Tesla C2075.

But when I ran 10 million mesh model, the following message log is shown.

 

Q1: Why GPU device shown in analysis log has changed ?

Q2: Is there a way to know the necessary memory size?

       Accourding to system log monitor of Windows, max private byte was 20GB.

 

 

Matrix Solver Settings:
======================
GPU device = Quadro 2000

 

** WARNING 302151 ** The GPU card memory is not large enough to run the full
                     model.  Therefore, the GPU card will only be used for some
                     time steps when the model is partially filled.

 

Thanks,

 

Message 6 of 8
scd310
in reply to: scd310

How much time did it take to complete the analysis? How were the results, i mean more accurate?

 

Thanks.

 

-

Samarth

Message 7 of 8
tim.vanast
in reply to: fukushige

The 10 million element model will not be able to fully solve on your GPU without running out of memory, while the 1.7 million element model will fully solve on the GPU alone.  Your results will be the same if you use a GPU or not.  So the 10 million element model will solve part of the analysis on the GPU, then switch over to only CPU when the GPU has used all of its memory.

 

Tim

Message 8 of 8

Hi,

 

Don't remember how to handle, but please notice that your AMI is using for CUDA de FX2000 and not the Tesla. It looks like your Tesla is waiting for action. Anyway I'm not sure that the Tesla will hold the 10 million elements.

 

Regards

Can't find what you're looking for? Ask the community or share your knowledge.

Post to forums  

Autodesk Design & Make Report