benchmarking_examples

Installation

• have CoreNeuron installed
• cd src then make clean then make
• prerequsites g++ > 7.5.0 , Cuda > 10.1.3.1 (I used 11.1)
• this will give you neuroGPU executable in benchmarking_examples/bin
• other prequsites
• pip install -r requirements.txt

Setup

• there are many different way to launch neuroGPU and core_neuron. In order to configure these experiments you'll need to edit two files:
  • launch_neuroGPU.sh
  • launch core_neuron.sh
• you can see that I had these configured to use a complex version of Srun due to some problems linking core_neuron with the local version of python on CORI.
• you'll want to delete that and add something of the order mpiexec -n <num_process> x86_64/special -mpi -python core_neuron_example.py or mpiexec -n <num_process> nrniv -mpi -python neuroGPU_example.py.
• $num_process and $cores will be specific to your machine.

Launching Experiments

• Set method in experiment_design.txt to be core_neuron or neuroGPU and then do sh launch_experiments.sh. Note, this will overwrite my benchmarking results in neuroGPU/logs or core_neuron/logs.
  • you can either move those logs or just check them out from main branch if you'd still like to see them.

Gathering results

• just run python collect_experiments.py to parse the logs.