Could you tell me how much the calculation time can be shortened?

General Questions
1

Right now I’m using a personal computer with 6 cores.
In the future, I am considering purchasing a 16-core personal computer.
Could you tell me how much the calculation time can be shortened?

2

It’s impossible to give specific numbers without discussing a particular case. Scalability depends on the solver (PaStiX, Pardiso or Spooles in the case of CalculiX) and given model (analysis type, mesh size and so on). That’s why different benchmarks are used. Anyway, don’t expect that the speed-up will be linear.

3

I would focus more on a high base/maximum speed GHz and memory than number of cores. Seems CalculiX doesn’t take too much advantage of more than 8 cores. You have some reference values on computation times with CalculiX at this post:

Check last Ryzen™ 9 5900X or HX numbers.

4

¿Don’t you think it could be interesting to have a Benchmark post/file also for Prepomax?. ¿A .pmx file that could be a reference for the users to compare different hardware configurations and the most important, to check their solver set up and its performance compared with similar hardware?

5

Here you can find some benchmark data based on the examples by M. Kraska: CalculiX solver

6

I should said: “Seems CalculiX doesn’t take too much advantage of more than 8 processors”

7

Hi,

I have imported the reference file. Here the original:

https://www.tu.kielce.pl/~rokach/instr/misc/solvers_test.7z

Seems BC and *.inp file is sliced in several files that are called later with an *Include card.

I have cut and paste everything into the same inp file and import from Prepomax but the import fails.

11/13/22 18:26:06 Set new model properties: Model space: 3D, Unit system: mm, ton, s, °C
11/13/22 18:26:10 Import file: \PREPOMAX\Benchmark\solid_begin.inp
11/13/22 18:26:10 Reading keyword line: *NODE, NSET=Nall
11/13/22 18:26:11 Reading keyword line: *ELEMENT, TYPE=C3D8R, ELSET=Eall
11/13/22 18:26:11 Reading keyword line: *NODE, NSET=Nall
11/13/22 18:26:11 Reading keyword line: *ELEMENT, TYPE=C3D8R, ELSET=Eall
11/13/22 18:26:11 Reading keyword line: *MATERIAL, name=steel
11/13/22 18:26:11 Reading keyword line: *solid section, elset=Eall, material=steel
11/13/22 18:26:11 Reading keyword line: *BOUNDARY
11/13/22 18:26:11 Reading keyword line: *STEP
11/13/22 18:26:12
11/13/22 18:26:12 Error: *end step
11/13/22 18:26:12 There were errors while importing the file/files.

¿Is there a way to fix this for everybody to have one and the same Benchmark.pmx file as reference?

Here is what I tried:

8

It’s better to recreate it based on the geometry. Especially since this way one will be able to change the mesh density (however only tetrahedrons can be generated in PrePoMax).

Benchmark.pmx (2.6 MB)

9

It can help me and other users to detect wrong configurations or points where the actual set up could be improved to increase the speed.
File Benchmark.pmx. Maximum element size adjusted to 4mm.

11/13/22 21:42:00 Nodes: 82592
11/13/22 21:42:00 Elements: 54076

screenshot.89

Win10
16 Gb (15,4 Usable)
Nº of CPU Cores 8
Nº of Threads 16
Base Clock 3.3GHz
Max. Boost Clock Up to 4.6GHz

Best runs:

CalculiX Version 2.19
Pardiso_MKL
True CCX_NPROC_STIFFNESS 12
True OMP_NUM_THREADS 4
True CCX_NPROC_RESULTS 12
Total CalculiX Time: 15.57s

CalculiX Version 2.20.
Pastix (without mixed precision)
True CCX_NPROC_STIFFNESS 6
True OMP_NUM_THREADS 6
True CCX_NPROC_RESULTS 12
Total CalculiX Time: 14.11