Cadence simulations on the HPC

From Applied Optics Wiki
Jump to: navigation, search

This is a work in progress document noting some of the issues with running Cadence simulations on the HPC. The point is to get very large simulations running that aren't tractable on the machines we use as desktops. Essentially we're just using it as a single computer that has more processors and memory than normal. We aren't exploiting the parallel nature of the HPC.

==Configuration==Setup->High-Performance Simulation... Click "APS". Also choose an accuracy override if you wish.

In a transient simulation go to the transient options windows. Click the Output tab and uncheck "Save Final Op Pt". If you are simulating a large circuit this can take a long time to write to disk.

Submission

Options are:

#$ -V -cwd
#$ -hard -l h_rt=18000
#$ -pe smp 1

Choose h_rt (max runtime) carefully - make it an amount longer than you anticipate the simulation to run, but don't go crazy. Shorter jobs are more likely to move up the queue quickly.

If you need >16GB RAM, use:

#$ -hard -l bigmem=true,h_rt=18000