To achieve your goal—running two independent tasks on a Slurm node with only 2 CPU cores, but allowing one task to start immediately if only 1 core is available—you need to adjust your resource request to avoid the strict "all-or-nothing" behavior of --cpus-per-task and --ntasks.

Problem with Your Current Command

Your command:

bash
Copy
srun --cpus-per-task=1 --ntasks=2 bash -c 'echo $SLURM_PROCID'

• Requests 2 tasks, each with 1 CPU core.
• Slurm will wait until 2 cores are free before launching both tasks, even if they are independent.

---

Solution: Use --ntasks-per-core

To allow one task to start immediately if only 1 core is available, use:

bash
Copy
srun --ntasks=2 --cpus-per-task=1 --ntasks-per-core=1 bash -c 'echo $SLURM_PROCID'

• --ntasks-per-core=1 tells Slurm to pack tasks as tightly as possible on available cores.
• This allows one task to start if only 1 core is free, and the second task will start when another core becomes available.

---

Alternative: Use --overcommit

If you want to force Slurm to start tasks even if resources are oversubscribed (not recommended for CPU-bound tasks, but useful for I/O-bound or lightweight tasks):

bash
Copy
srun --ntasks=2 --cpus-per-task=1 --overcommit bash -c 'echo $SLURM_PROCID'

• --overcommit allows Slurm to ignore strict resource limits and start tasks immediately, even if it means oversubscribing cores.

---

Key Takeaways

Recommendation: Use --ntasks-per-core=1 for your case. It respects Slurm's resource management while allowing tasks to start as soon as cores are free.