CLI
The gpu-tracker
command-line interface allows tracking computational-resource-usage of an arbitrary shell command.
For example, one may want to profile a command that runs a script or a command ran in a high-performance-computing job.
Below is the help message shown from gpu-tracker --help
.
See the CLI section of the Tutorial for examples of using the CLI.
Usage:
gpu-tracker -h | --help
gpu-tracker -v | --version
gpu-tracker --execute=<command> [--output=<output>] [--format=<format>] [--st=<sleep-time>] [--ru=<ram-unit>] [--gru=<gpu-ram-unit>] [--tu=<time-unit>] [--nec=<num-cores>] [--guuids=<gpu-uuids>] [--disable-logs] [--gb=<gpu-brand>]
Options:
-h --help Show this help message and exit.
-v --version Show package version and exit.
-e --execute=<command> The command to run along with its arguments all within quotes e.g. "ls -l -a".
-o --output=<output> File path to store the computational-resource-usage measurements. If not set, prints measurements to the screen.
-f --format=<format> File format of the output. Either 'json' or 'text'. Defaults to 'text'.
--st=<sleep-time> The number of seconds to sleep in between usage-collection iterations.
--ru=<ram-unit> One of 'bytes', 'kilobytes', 'megabytes', 'gigabytes', or 'terabytes'.
--gru=<gpu-ram-unit> One of 'bytes', 'kilobytes', 'megabytes', 'gigabytes', or 'terabytes'.
--tu=<time-unit> One of 'seconds', 'minutes', 'hours', or 'days'.
--nec=<num-cores> The number of cores expected to be used. Defaults to the number of cores in the entire operating system.
--guuids=<gpu-uuids> Comma separated list of the UUIDs of the GPUs for which to track utilization e.g. gpu-uuid1,gpu-uuid2,etc. Defaults to all the GPUs in the system.
--disable-logs If set, warnings are suppressed during tracking. Otherwise, the Tracker logs warnings as usual.
--gb=<gpu-brand> The brand of GPU to profile. Valid values are nvidia and amd. Defaults to the brand of GPU detected in the system, checking NVIDIA first.