ntasks

Ntasks is a parameter used in parallel computing to specify the number of tasks (often MPI processes) to be launched for a job. It is commonly exposed by batch schedulers and MPI runtimes to control the scale of a parallel execution.

In Slurm, the option --ntasks (short -n) requests that a total of N tasks be started across the allocated nodes. This typically maps to MPI ranks or other worker units used by the application. The value interacts with other resource specifications such as --nodes, --cpus-per-task, and, in some configurations, --ntasks-per-node, which caps the number of tasks that can run on each node. An example: sbatch --ntasks=64 myjob.sh or srun -n 64 ./my_program.

Ntasks is conceptually distinct from the number of CPUs or cores allocated. For example, you might allocate several nodes with a given number of CPUs per task and set --ntasks=32, depending on the launcher, to achieve a desired level of parallelism. The actual mapping from tasks to hardware depends on the scheduler and runtime being used.

Notes: In MPI environments, ntasks often correspond to MPI ranks; in other setups it may refer to independent worker processes. The term is used across batch systems such as Slurm, Torque/PBS, and others, sometimes with different command-line flags but preserving the same general meaning.