Search code examples
pythoncpudistributed-computingray

Limiting CPU resources of Ray

I'm trying to manage the resources of a remote machine that we use for a daily task (that uses Ray). Is it possible to limit the number of CPUs (or equivalently the number of workers) that Ray uses?

The remote machine has 16 cores. Can I limit Ray to use only 12 of them or so?

Solution

  • You can limit resources using:

    # To start a head node.
ray start --head --num-cpus=12

# To start a non-head node.
ray start --redis-address=<redis-address> --num-cpus=12

    Or via ray.init:

    ray.init(num_cpus=12)

    Source: Ray documentation.