Resource allocation #

The docker container run command provides options for allocating compute resources.

GPU allocation #

You can specify the GPUs to be used. By default (no specification), no GPUs are allocated. Please check the device IDs by running the nvidia-smi command on the server. For example, igpu01 has four GPUs, each assigned a device ID from 0 to 4.

To avoid resource conflicts, please use device IDs to specify GPUs. Use --gpus all only if you want to use all GPUs.

--gpus [Device ID of the GPU to be used / all]
# e.g. Single GPU (device 0 is specified)
--gpus '"device=0"'
# e.g. Multi-GPUs (specify device 0 and 2)
--gpus '"device=0,2"'
# e.g. When using all GPUs
--gpus all

Shared memory allocation #

Shared memory is used for pre-processing and other processes that do not use the GPUs. The default (no specification) allocates only 64 MB, which often results in insufficient memory, so it is recommended to specify this.

--shm-size [memory size]
# e.g.
--shm-size 16gb

CPU allocation #

You can allocate the required amount of CPUs. The default (unspecified) will use all CPUs.

--cpus [amount of CPUs]
# e.g.
--cpus 2

Timeout time #

You can set a timeout period (in seconds) before the container is stopped.

--stop-timeout [timeout time]
# e.g. Stops in 1 hour
--stop-timeout 3600