Relion

Example

Basic usageLaunch from terminalSubmission script (save as relion5_slurm.sh)Queue submission (CPU)Queue submission (GPU)

To use Relion version 5, select 5.0 from Relion_Testing in Open OnDemand

For most cases, Relion GUI doesn't need gpu. Please start Relion on a CPU node, and then submit jobs to the GPU partitions through Relion GUI.

        $ ssh -X username@spartan.hpc.unimelb.edu.au    # login to spartan
        $ sinteractive --x11 --time=06:00:00 --cpus-per-task=4  # start an interactive job
        $ module load intel-compilers/2022.1.0      # load dependencies
        $ module load Relion/5.0                # load relion

        $ cd /data/gpfs/projects/punim????/ProjectDirectory # navigate to your RELION project directory
        $ relion                        # open the relion GUI

        #!/bin/bash

        #SBATCH --ntasks=XXXmpinodesXXX
        #SBATCH --partition=XXXqueueXXX
        #SBATCH --qos=XXXextra4XXX
        #SBATCH --gres=gpu:XXXextra1XXX
        #SBATCH --cpus-per-task=XXXthreadsXXX
        #SBATCH --time=XXXextra2XXX
        #SBATCH --mem-per-cpu=XXXextra3XXX
        #SBATCH --error=XXXerrfileXXX
        #SBATCH --output=XXXoutfileXXX
        #SBATCH --tmp=300G 

        #INFO
        echo "Starting at `date`"
        echo "Running on hosts:$SLURM_NODELIST"
        echo "Running on $SLURM_NNODES nodes."

        srun XXXcommandXXX

RUNNING:
Number of MPI procs:            12      # between 1-12 should be fine for most jobs
Submit to queue:            Yes     
Queue name:                 cascade
Queue submit command:           sbatch
Number of gpus:             0
Wall time:              20:00       # days-hours:min:secs - shorter will run sooner in the queue
Memory:                 12G     # per CPU. total = mem * MPI - lower will run sooner in the queue
QoS:                    normal
Standard submission script:     /path/to/relion5_slurm.sh
Minimum dedicated cores per node:   1       # leave as default

COMPUTE:
Use parallel disc I/O?          Yes
Number of pooled particles:     3
Skip padding?               No
Pre-read all particles into RAM?    No
Copy particles to scratch directory:    /tmp/       # temporarily transfers data to SSD faster access during the run
Combine iterations through disc?    No
Use GPU acceleration?           Yes
Which GPUs to use:          # leave blank

RUNNING:
Number of MPI procs:            5       # Works best as the number of GPUs+1 
Number of threads:          6       # MPI*thread = total CPUs (gpu-a100 has max 31)
Submit to queue:            Yes     
Queue name:                 gpu-a100
Queue submit command:           sbatch
Number of gpus:             4       # fewer will run sooner in the queue
Wall time:              ?-??:??:??  # shorter will run sooner in the queue
Memory:                 8G      # per CPU. Total = mem*MPI*threads, max 495G (e.g. 16G for 30 CPU) - lower will run sooner in the queue
QoS:                    normal
Standard submission script:     /path/to/relion5_slurm.sh
Minimum dedicated cores per node:   1       # leave as default