solution speed and number of nodes

[kmguleren]

Hello,

The first impression on this code is quite upsetting, since we could not run the code with more than 26 processors and moreover it runs quite slowly like it does not have any iterative methods to solve the discretized equations. For example, it completed 100 iterations with 26 high performance processors (Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz) for a 2d mesh of 692 cells in 120 seconds. Is there anything that we are doing crucially something wrong, or have you get any similar complaints?

I would really appreciate if you could provide some help on using this code.

Best regards and many thanks,

Assoc. Prof. Dr. Kürşad Melih Güleren Anadolu University

[mlopez14]

Sorry to hear you're not getting the performance you expected. It is worth considering two items:

1. So few elements per core can be making the MPI communications the bottleneck. Have you tried with fewer cores? Can you confirm your cluster's setup is operating normally?
2. There is an annoying time penalty paid at the beginning of each simulation. Before any time-steps are taken, many geometric parameters are calculated for each element. Are the 120 seconds counted starting from the first time-step or is this a total including the setup time?

The code is designed with GPU computing as a priority: explicit time-stepping leverages the GPU's hyper-parallel, memory-limited architecture best.