Learning Forum

[Ruzan]

Hi everyone,

I have a question regarding using Lumerical hpc engine licenses. My current understanding is that if we run simulations standard Lumerical engine, it should be efficiently using only 32 cores, and, in principle, increasing the number of cores should yield no benefits. However, the simulation time still significantly decreases if I use a larger number of cores. Why is it the case? 

Also, does anyone have experience using hpc Lumerical licenses? Did you see clear benefits as compared with just using standard engine licenses?

Thanks!

[Lito Yap]

@Ruzan, 

Academic licenses does not work with Ansys HPC licenses. The HPC license is only available for commercial/enterprise licenses.

Based on your current entitlement/licenses, you should be able to run a single simulation up to 1280 processes/cores. The simulation/project and the machine running the simulation will affect the performance. You can increase the number of processes/cores used to the run the simulation up to a point that increasing the number of processes/cores will not improve the performance. 

We have the following KB that might help:
FDTD Performance Benchmarks – Ansys Optics 
Determining the optimum resources to run FDTD – Ansys Optics

[Ruzan]

Thank you for your response. Just a clarification: do you mean 1280 processes or 128? And when running 128 processes, will I be utilizing 4 engine licenses?

I am curious whether the hpc license enables better utilization of computer resources. Your benchmark simulations show that the simulation time may actually increase when adding additional cores. But are you able to avoid this when using hpc licenses?

Also, do you think the hpc licenses will be available to academic researchers in the nearest future?

 

Thanks!

 

 

[Lito Yap]

@Ruzan, 

Based on your current entitlements/licenses, 2x (5 tasks) Ansys Lumerical Academic Research license you have: 

2×5 (10) GUI licenses for each Ansys Lumerical Product 
2x5x4 (40) solve licenses for each Ansys Lumerical Product/Solver
(i.e. 40x FDTD solve, etc.)
So you can run a single (1) simulation up to a maximum of (40×32) 1280 processes/cores 
>>> List of licensed features by product – Ansys Optics 

The enterprise license is only available for commercial accounts I don't think this will be available for academic accounts. Check with your Ansys account manager for pricing and availability. The license does not affect the performance of the application. It only provides access to the application and the additional license (solve or HPC license) simply allows the application to run with more processes/cores.

The performance is dependent on the simulation/design and the machine running the simulation. Whatever license you are using standard or enterprise, once the bandwidth limitation of the machine is reached it will not make the simulation run faster even if you increase the number of processes/cores. Or when the simulation is overspliced (running with more cores) the performance will be slower.

e.g. running a small/simple simulation like the “Microwire polarizer” at 96 processes would be slower compared to running it at 32 or less processes.