Cluster 7

From DISI
Jump to navigation Jump to search

Introduction

A cluster built on Rocky 9 linux distribution which is under the RHEL umbrella.

How to Request for Access

Contact a system administrator.

How to Login

  1. Remote
    ssh <user>@epsilon.compbio.ucsf.edu

SLURM Nodes

  • To list partitions
    sinfo
  • To list all nodes and their information
    sinfo -lNe

CPU Servers

  • cpu02 (128 Cores, 256 Threads, 1TB RAM)
  • cpu[03-17] (48 Cores, 96 Threads, 256GB RAM)
  • I make that 1696 concurrent processes.

GPU Servers

  • gpu01 8 x GeForce RTX 2080 Ti 10 GB VRAM. (48 Cores, 96 Threads, 758GB RAM, 8 x RTX 2080 Ti)
  • gpu02 4 x GForce RTX 3090 . 24 GB VRAM. (48 Cores, 96 Threads, 758GB RAM, 4 x RTX 3090). 2.4 Ghz. Xeon Gold 6240R.
  • gpu03 5 x GeForce GTX 1080 Ti 10 GB VRAM + 3 x Titan Xp 12 GB VRAM. 64 cpus. Xeon E5-2697 v4 @ 2.6 GHz
  • gpu04 4 x Titan RTX. 24 GB VRAM. 96 cores. Xeon Gold 6240R @ 2.4 GHz
  • gpu05 4 x Titan RTX. 24GB VRAM. 96 cpus. Xeon Gold 6240R 2.4 GHz
  • gpu06 4 x GeForce RTX 3090. 24 GB VRAM. 96 cpus. Xeon Gold 6240R 2.4 GHz
  • I make that 32 GPUs and 544 cpu processes.

NFS Servers

  • home01
  • hdd02

Global Modules/Software

  • To check the list of available modules
    /* Long version */
    module available 
    
    /* OR */
    
    /* Short version */
    ml av
  • To load a module(s)
    /* Single module */
    module load dock 
    
    /* OR */
    
    ml dock 
    
    ==================================
    
    /* Multi-module */
    module load dock python/3.12, schrodinger
    
    /* OR */
    
    ml dock python/3.12 schrodinger
  • To list all loaded modules
    module load
    
    /* OR */
    
    ml
  • To unload a module
    module unload <module>
    /* OR */
    ml unload <module>
  • To unload all modules
    module purge 
    /* OR */
    ml purge


Guidelines and Recommendations

  • /nfs/home is for code (backed up to github or gitlab)
  • /nfs/hdd01/work/<yourid> and /nfs/hdd02/work/<yourid> are good places to keep docking jobs. write to your sysadmin to set up a folder for you.
  • jobs under slurm should write to local /scratch during the job, move back to your work area and clean up when job ends