Useful links

Tips / advice

Slurm job configuration tip

  • Memory: Each compute node has at least 2.5GB per core

    • Probably a safe amount of memory to request

  • Cores: Compute nodes typically have cores in multiples of 8

    • Requesting more cores may take longer to fulfill your request

    • It may help to understand when you reach diminishing returns with resources

    • Online rule of thumb suggests to stop adding cores once you only get 30% speed improvement from doubling cores

    • Maximum core count in the shared pool is 48 cores, but if you request that, it may take a while for your job to be picked up. Might be worth requesting fewer cores for a longer time

  • Time

    • Maximum run time is 14 days

      • Do not set your run time at 14 days, apparently it messes with the scheduler

    • Pick a time longer than you need, but not so long that the scheduler gives it very low priority