Ray actor only uses one core on a cluster managed using SLURM

I am using Ray on a cluster managed using SLURM and the Ray actor just uses one core even though I have allocated more cores to it.

Here is the python code:

import numpy as np
import ray
import time
import os

@ray.remote(num_cpus = 12)
class CPUActor(object):
    def __init__(self, n):
        self.A = np.random.randn(n, n)
        self.B = np.random.randn(n, n)

    def mul(self):
        C = np.matmul(self.A, self.B)
        return np.mean(C)

n = 5000
ray.init(address = 'auto')
actor = CPUActor.remote(n)

start_time = time.time()
K = 10
for i in range(K):
    temp = ray.get(actor.mul.remote())

used_time = time.time() - start_time
print(f"used time in ray: {used_time:.4f}", flush = True)

This is the slurm script I used to run the Python code:

#SBATCH --job-name=test_ray
#SBATCH --partition=xeon-p8
#SBATCH --time=00:10:00
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=1
#SBATCH --cpus-per-task=36
#SBATCH --open-mode=truncate
#SBATCH --output=./use_ray.txt

## get node names
nodelist=$(scontrol show hostnames $SLURM_JOB_NODELIST)
echo "node list: $nodelist"

srun --nodes=1 --ntasks=1 --nodelist=${nodes_array[0]} \
     --output=./ip_address_${SLURM_JOB_ID}.txt \
     --open-mode=truncate \
     --error=/dev/null \
     hostname --ip-address

ip_prefix=$(cat ./ip_address_${SLURM_JOB_ID}.txt) # making redis-address

export ip_head
echo "ip_head: ${ip_head}"

echo "STARTING HEAD at ${nodes_array[0]}"
echo "num of cpus:", $SLURM_CPUS_PER_TASK
srun --nodes=1 --ntasks=1 --nodelist=${nodes_array[0]} \
     ray start --head --block \
     --port 6379 --temp-dir=/home/gridsan/dingxq/tmp/ray \
     --num-cpus=$SLURM_CPUS_PER_TASK &

sleep 10

export RAY_ADDRESS=$ip_head

python ./use_ray.py


@Xinqiang_Ding In Ray, an actor is running in a single thread so it could only use at most one core; unless you implement multiple threads inside the actor, or using Threaded Actors. You could try to create two actors which should be able to utilize multiple cores.