CompG Details and Etiquette
Best practices for submitting jobs to compG, excluding compG000 and compG001.
Note that if you intend to, or are using compG, it is strongly advised that you join the BMRC GPU user mailing list: rescomp-users-gpu@well.ox.ac.uk. Simply email a request to be included to bmrc-help@medsci.ox.ac.uk.
Much of the following information has been taken from the bmrc website (see below).
Available GPU Resources
Information taken from the bmrc website on 24/10/2019.
Server compG00? | GPUtype | Number of GPUs | GPU memory (GB) | Cuda Version | OS | Role | Access |
0 | k80 | 8 | 12 | 8+ | CentOS 6.8 | Strubi cryosparc jobs | open |
1 | TitanX | 1 | 12 | 8+ | Strubi cryosparc jobs | restricted | |
2 | v100 | 1 | 16 | 9+ | currently offline | open | |
3 | p100 | 4 | 16 | 8+ | CentOS 6.8 | Queued GPU jobs | open |
4 | p100 | 4 | 16 | 8+ | CentOS 6.8 | Queued GPU jobs | open |
5 | 1080Ti | 4 | 12 | 8+ | Prototyping and development | open | |
6 | 1080Ti | 4 | 12 | 8+ | Prototyping and development | open | |
7 | 1080Ti | 4 | 12 | 8+ | Prototyping and development | open | |
8 | 1080Ti | 4 | 12 | 8+ | Prototyping and development | open | |
9 | p100-os7 | 4 | CentOS 7.3 | Queued GPU jobs | open | ||
10 | p100-os7 | 4 | CentOS 7.3 | Queued GPU jobs | restricted | ||
11 | p100-os7 | 4 | CentOS 7.3 | Queued GPU jobs | restricted |
Note that users can use NVidia DGX-1V (8 x V100) hosted by ARC, run by the IT services upon request. Contact bmrc-help@medsci.ox.ac.uk.
Submitting Jobs
The 1080Ti cards are gaming cards that are a bit slower and cheaper than the HPC-grade P100 cards. To make access as fair as possible, please submit any large jobs to the queue system.
#################EXAMPLE SUBMISSION SCRIPT. Many thanks to Rob Stass ###################
#!/bin/bash
# Name of the job
#$ -N motioncor2_job
# Job log file (optional)
#$ -o motioncor2-$JOB_ID.log
#queue information
#$ -cwd -V
#$ -j y
#$ -P strubi.prjc
#$ -q gpu8.q
#$ -l gpu=1
#$ -l gputype=p100
# or can request gputype=p100
#load cuda
module unload cuda
#If you want to use an older version of motioncor2 uncomment/comment the lines to load cuda/7.5 instead.
#module load cuda/7.5
module load cuda/8.0
num_gpu=`echo $SGE_HGR_gpu | wc -w`
gpu_identifiers=`seq -s " " 0 1 $(($num_gpu-1))`
/apps/strubi/motioncorr/20170130/MotionCor2-01-30-2017 -Gpu $gpu_identifiers <other_motioncor2_parameters>
#############################################################################
For more information head to:
Note that the following website may only be accessed via your rescomp login credentials.
https://help.bmrc.ox.ac.uk/tutorials
https://rescomp.well.ox.ac.uk/wiki/Using_the_Rescomp_GPU_resources