"Or maybe this limit is what the job max. needs and there is no need to go higher."

Probably yes. Yu can set all limits in the env-file; the required memory will also be printed in the dat-file.

"only 25 % is used. Is there anyway to increase this?"

Ar those dual/quad core processors? In this case, increase the number of processors Abaqus uses (cpus=2, 4 or whatever is required).

"For example for a model I am using tetrahedral elements, and it seems to take longer."

Depends on what kind of calculation you are doing - in implicit, this may be due to wavefront ordering; in explixit, it may be due to tets having smaller dimensions and reducing the minimum time increment.

Thank you very much Mr. Bäcker. 

I checked the dat file and it was there abour 30 Gb! So that is the reason which I think depends on DOFs.

I also had applied the changes in env. file when I wanted to submit the 2. job as:

.....

memory = "90 %"

cpus=2

import driverUtils, os 

....

the last line I put to show where I added the lines. I hope it is the correct way. Anyway as I mentioned I also requested more memory in the submission line of the job.

-------

There are 2 nodes, each node has 4 CPU cores or 8 logical processors. (according to this link

http://www.fixedbyvonnie.com/2014/08/ever-wonder-set-affinity-right-click-process-task-m/#.WEAxXbIrKUk)

How can I know how many cores is optimal to use?

Best regards

AFAIK, abaqus counts each logical processor as a "cpu", therefore you may want to use cpus=8, for example.

In most models (if they are sufficiently large), Abaqus scales well, so more cpus give you less time (but require more license tokens).

thank you very much.

I have 2x8 logical processors, hence I could maybe also write cpus=16.

maybe you mean cpu cores.

But implementing cpus=8

I get a higher percentage of cpu usage. (see the image please, the 2 digit number is the cpu usage percentage, and the other is the allocated RAM)

I do not know if it is better to run one simulation, or parallel more simulations? 

The parallelization is heavily dependent upon geometry, e.g. the one-dimensional explicit dynamic problem takes on my workstation with 4xXeons about 10 sec at single logical processor (core) and about 240 sec at 64 logical processors (see Project "1D problem: wave propagation in a bimodular rod), at the same time 2D and 3D problems can be successively parallelized

Hi Vahid Ebrahimzade Isfahani,

Please also let us know if you have found an optimum setting to speed up with the computing power you have.

Is it okk for purchasing abacus with 1 core only ?

1 core or 4 core, which one do you suggest and why?

Good question.. Please share me the best answer might you trust...

Regards…

Using tet elements will consequently increase the number of elements hence is always better to use hex elements. Of course, it is inevitable if the geometry is too complex. Adding to these if your elements are quadratic, then it is no surprise that it takes longer time for your simulation (10 nodes per element). You can use mass scaling to speed up, if you are conducting simulation on Abaqus/explicit. Also try using an optimum number of CPUs, the communication overhead increases way higher if you unnecessarily use large number of CPUs.

Thank you

Hello @Farshad Abbasi,

It is good practice to also give the source of the information that you have provided, as the newsletters that Simuleon release for free are useful for everyone.

Source of the 7 good practices - https://info.simuleon.com/blog/7-tips-to-help-abaqus-run-faster

Hi Martin Baeker ,

Although I have decreased the number of elements in my model by reducing the size of geometry, the simulation requires more memory than the bigger geometry with more number of elements.

All elements in both models are linear.

Have you come across this problem?

No, and I think that is a bit surprising. You may find some information on memory usage in the dat-file, that might help you to understand this.

sometimes, your mesh discretization and the interaction between different materials or models can affect the simulation speed, because of that make sure your mesh is good (node-to-node interaction).

Chandra Sajana You maybe right. I have acoustic elements and 3-D elements in my model. I am using tet elements for both acoustic and solid.

At acoustic and solid element interaction, the element sizes are following:

acoustic elements: 1

3-D elements: 0.25

How to verify for good node-to-node interaction?

Dear Ebrahimzade

I think you should use multiple processors (8 or more) in the Parallelization module. Also, there is no need to use the Full option in nodal output precision. Furthermore, you can reduce the number of elements or utilize the hexahedral elements instead of tetrahedral.

Hello Dr. Baeker,

Is there a limit to how many processors I can use in ABAQUS? The job runs with a maximum of 12 cpus. For value more than this, it does not run.

Regards,

Nidhish