======GPU Nodes======

**Upgraded: more GPUs added.**

The Lengau cluster at the CHPC includes 9 GPU compute nodes with a total of 30 Nvidia V100 GPU devices.  There are 6 ''gpu200''//n// nodes with 3 GPUs in each, and 3 ''gpu400''//n// nodes with 4 GPUs in each.


^  GPU Node  ^  CPU Cores  ^  GPU Devices  ^  Interface  ^
| ''gpu2001''  |  36 | 3× Nvidia V100 16GB  | PCIe  |
| ''gpu2002''  |  36 | 3× Nvidia V100 16GB  | PCIe  |
| ''gpu2003''  |  36 | 3× Nvidia V100 16GB  | PCIe  |
| ''gpu2004''  |  36 | 3× Nvidia V100 16GB  | PCIe  |
| ''gpu2005''  |  36 | 3× Nvidia V100 **32GB**  | PCIe  |
| ''gpu2006''  |  36 | 3× Nvidia V100 **32GB**  | PCIe  |
| ''gpu4001''  |  40 | 4× Nvidia V100 16GB  | **NVlink**  |
| ''gpu4002''  |  40 | 4× Nvidia V100 16GB  | **NVlink**  |
| ''gpu4003''  |  40 | 4× Nvidia V100 16GB  | **NVlink**  |

> Jobs that require 1, 2 or 3 GPUs can be allocated to any node, and will share the node if the job does not use all the GPU devices on that node.  Jobs that require 4 GPUs can only be allocated to ''gpu4*'' nodes and will not be shared, obviously.





=====Policies=====

====Access====

Principle Investigators apply for GPU access for their Research Programme members through the CHPC [[https://users.chpc.ac.za/helpdesk/|Helpdesk]].  RP members may not apply directly. E-mailed applications will not be considered. 

====Allocation=====

Research programme allocations will be depleted by a factor of the wallclock time and the number of GPUs (1, 2, or 4) requested by the job.

  gpu_allocation_used = 40 * runtime * ngpus

======Usage======

=====GPU applications on Lengau=====

Some pre-built applications have automated scripts that you can use to launch them:

  * [[howto:gromacs|GROMACS]]
  * [[howto:AMBER|AMBER18]]
  * [[howto:NAMD|NAMD]]

=====GPU Job Scripts=====

====GPU Queues====

There are four queues available in PBSPro which access the GPU nodes:

^  Queue name  ^  Max. CPUs  ^  Max. GPUs  ^  PBSPro options  ^ Comments  ^
|  **gpu_1**  |  9 |  1 | ''-q gpu_1''\\ ''-l select=1:ncpus=9:ngpus=1''  | Access one GPU device only per job.  |
|  **gpu_2**  |  18 |  2 | ''-q gpu_2''\\ ''-l select=1:ncpus=18:ngpus=2''  | Access two GPU devices per job.  |
|  **gpu_3**  |  36 |  3 | ''-q gpu_3''\\ ''-l select=1:ncpus=30:ngpus=3''  | Access three GPU devices per job.  |
|  **gpu_4**  |  40 |  4 | ''-q gpu_4''\\ ''-l select=1:ncpus=40:ngpus=4''  | Access four GPU devices on NVLink nodes.  |

Note the ''ncpus'' parameters above is the maximum that should be set to match the number of GPU devices you need.

====GPU Queue Limits====

The maximum wall clock time on all GPU queues is 12 hours.
<code bash>
#PBS -l walltime=12:00:00
</code>

> It is better to specify a shorter walltime if your code executes in less time: this allows the scheduler a better chance of running your job sooner.

====Interactive Job on a GPU Node====

A single //interactive// session may be request on a GPU node by

<code bash>
qsub -I -q gpu_1 -P PRJT1234 -l select=1:ncpus=9:ngpus=1
</code>
**NB:** Replace //''PRJT1234''// with **//your//** project number.

The default time for an interactive session is 1 hour.

====Example Job Script====

<code bash>
#!/bin/bash
#PBS -N nameyourjob
#PBS -q gpu_1
#PBS -l select=1:ncpus=4:ngpus=1
#PBS -P PRJT1234
#PBS -l walltime=4:00:00
#PBS -m abe
#PBS -M your.email@address
 
cd /mnt/lustre/users/USERNAME/cuda_test

echo
echo `date`: executing CUDA job on host ${HOSTNAME}
echo
echo Available GPU devices: $CUDA_VISIBLE_DEVICES
echo

# Run program
./hello_cuda
</code>

As usual, replace ''PRJT1234'' with your group's project name, ''your.email@address'' with your email address, and ''USERNAME'' with your cluster user name.

====GPU Memory====

Most of the V100 GPUs only have 16GiB of memory.  Two nodes each have three V100s with 32GiB of RAM.  To access these it is necessary to specify the exact node:

  #PBS -l select=1:host=gpu2005:ncpus=9:ngpus=1

or

  #PBS -l select=1:host=gpu2006:ncpus=9:ngpus=1

> Note that asking for a specific node is likely to lead to longer queue times as your job has to wait until that node becomes available.
  

=====Compiling GPU Code=====

The Nvidia V100 GPUs are programmed using the **CUDA** development tools.

To build a CUDA code (library or application) for the GPU nodes requires loading the appropriate CUDA module before compiling.    The current CUDA modules are:

<code>
chpc/cuda/11.2/PCIe/11.2
chpc/cuda/11.2/SXM2/11.2
chpc/cuda/11.5.1/PCIe/11.5.1
chpc/cuda/11.6/PCIe/11.6
chpc/cuda/11.6/SXM2/11.6
chpc/cuda/12.0/12.0
</code>

with version 12.0 the most recent version.

Note that the 11.x version modules are available in two types:  
    * the ''PCIe'' version is for PCIe bus nodes: ''gpu200x''
    * the ''SXM2'' version is for the SXM2 bus nodes: ''gpu400x''



====Further Reading====

  * Detailed [[https://www.olcf.ornl.gov/for-users/system-user-guides/summit/nvidia-v100-gpus/|overview]] of the V100.
  * Nvidia [[http://images.nvidia.com/content/volta-architecture/pdf/volta-architecture-whitepaper.pdf|whitepaper]] on the Volta architecture.
  * CUDA Toolkit [[https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html|Release Notes]]