Pyrosim simulation system performance

[Marziyeh Ohadi]

Hi everyone

I bought a computer with the specifications shown below for CFD modeling of fire with Pyrosim:

CPU: 2x Intel Xeon Platinum 8124M

36Core 72Thread 90MB Cache @ 3.0Ghz Base Clock

MB: Asrock EP2C621D16 WS

RAM: 256GB DDR4 ECC 2666

Storage: 1TB Samsung PM9A1 Nvme

Graphic Card:  Nvidia RTX A2000 12GB ECC

Power: MSI MPG1000 Full Modular

I also have this older computer with specifications shown below:

CPU: 11th Gen Intel(R) Core(TM) i7- 11700 @ 2.50GHz

8Core 16Thread 20MB Cache @ 2.5Ghz Base Clock

MB: ASUSTeK COMPUTER INC.

RAM: 32GB

Storage: Lexar 256GB SSD

WDC WD 10EZEX-00BBHA0

Graphic Card:  NVIDIA Geforce GTX 1650 

 I used both of them for the same model with about 1000000 cells sized 0.6 m. The only difference between them was the number of their computing networks. In the model with the old computer, I created 8 computing networks and assigned each to one core. And for the more powerful computer, 36 computing networks assigned each to one core

The result has been surprising for me. Although the new system is much stronger it took 2 hours longer for it to finish the simulation. It also has too many lags in loading 3D cad models compared with the older system.

Can anyone give me a clue what the problem is with this system?

[dr_jfloyd]

Do you have hyperthreading disabled on both? If it is disabled on the old computer but not the new computer, that could explain the differences.

If hyperthreading is disabled, do the computers have identical setups in terms of the operating system, number and version of drivers, etc? Even with hyperthreading off, 8 and 36 would be all the physical cores which means FDS will be sharing one or more cores with other software running on the computer.

Does the new machine have a proportionally larger memory bus for the increased number of cores? If cores are always waiting for access to memory, that will slow things down.

There could be some hardware issue with the new machine.

You may want to take your model and do a scaling study where you run the model with different numbers of cores on the new machine. 

[Marziyeh Ohadi]

Hi

thanks for your helpful answer.

Both systems hyperthreading had been enabled. I changed both of them to disable. The odd thing was having better performance in the new computer and not a satisfying performance in the old computer compared to previous results. Is it only helpful for Xeon CPUes?

--
You received this message because you are subscribed to the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fds-smv+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/bee6afd3-24ee-44cc-9fa7-77efd46b93e8n%40googlegroups.com.

[dr_jfloyd]

The general experience I have seen is that in most cases disabling hyperthreading improves performance. I have the 12th gen I9 in my laptop and disabling hyperthreading lead to improved performance when running FDS with MPI. Could be that for the 36 core XEON, that the performance hit is larger than for the 8 core I7. 

Another thing to check is if these are Windows machines is the power plan performance setting. The power saving plans can limit the number of active cores.