Command line options

Controlling where a simulation is executed

Any simulation in HOOMD can be run on the CPU or GPU. To control which, set the --mode option on the script command line. Valid settings are cpu and gpu.

hoomd some_script.hoomd --mode=cpu

When --mode is set to gpu and no other options are specified, HOOMD runs on the first GPU (number 0) in the system. This may not be desirable if another GPU in the system is faster or a simulation is already running on GPU 0 (HOOMD will run correctly with more than one simulation on a GPU as long as there is enough memory, but the performance penalty is severe). You can select which GPU to run on using the --gpu command line option.

hoomd some_script.hoomd --gpu=1

Note that specifying --gpu implies --mode=gpu. To find out which id is assigned to each GPU in your system, download the CUDA SDK for your system from http://www.nvidia.com/object/cuda_get.html and run the deviceQuery sample.

Multiple GPUs can be selected to run on by specifying a comma separated list:

hoomd some_script.hoomd --gpu=0,1

Any number of GPUs can be specified in the list.

If you run a script without any options

hoomd some_script.hoomd

HOOMD first checks if there are any GPUs in the system. If it finds one or more, it runs on GPU 0. If none are found, it runs on the CPU.

---

Controlling error checking on the GPU

Detailed error checking is off by default to enable the best performance. If you have trouble that appears to be caused by the failure of a calculation to run on the GPU, you can run with GPU error checking enabled to check for any errors returned by the GPU.

To do this, run the script with the --gpu_error_checking command line option:

hoomd some_script.hoomd --gpu_error_checking

---

You can always run

hoomd some_script.hoomd --help

to get a full list of the available command line options some of which may not be listed here.