Select a Processor Architecture

Choosing nodes for a job based on the processor micro-architecture.

Chinook currently has three types of standard compute nodes with differing Intel micro-architectures. It is possible to limit a job to run on just one or two processor types. This allows a binary built with more advanced instruction sets to be executed only on processors that support the instruction set. If a binary is not built for a specific instruction set, it will run on any of the three types.

The three micro-architectures which translate to a node type are:

To select a specific node type, use the "--constraint" option of the sbatch or srun commands. Multiple choices may be specified with the OR operator.

Example 1, request a job run only on Skylake/Cascade Lake nodes:

sbatch --constraint="Skylake" jobScript.slurm

Example 2, request a job run only on Broadwell or Skylake nodes:

sbatch --constraint="Broadwell|Skylake" jobScript.slurm

Example 3, request a job run only on a single node type, but either Broadwell or Skylake nodes may be used (aka matching OR):

sbatch --constraint="[Broadwell|Skylake]" jobScript.slurm

By adding a constraint, the number of possible nodes a job can run on will be reduced. This means it may take more time for the scheduler to make those nodes available and launch the job.

Last updated