Overview

Compute Resources

More detailed information on system resources can be found on our Compute Resources page.

Containers with GPU Support

Singularity containers are available as modules on HPC for GPU-supported workflows. For more information, see our documentation on Containers.

Accessing GPUs

Information on how to request GPUs using SLURM can be found in our SLURM Documentation.

Training

For a list of training resources related to GPU workflows, see our Training documentation.

Cluster Information

Puma

Puma has a different arrangement for GPU nodes than Ocelote and ElGato. Whereas the older clusters have one GPU per node, Puma has four. This has a financial advantage for providing GPU's with lower overall cost, and a technical advantage of allowing jobs that can use multiple GPU's to run faster than spanning multiple nodes.  This capability comes from using a newer operating system.  
Each node has four Nvidia V100S model GPUs. They are provisioned with 32GB memory compared to 16GB on the P100's.
Image Added

Ocelote

Ocelote has 45 compute nodes with Nvidia P100 GPUs that are available to researchers on campus.  There will be fairshare limitations but the intention is for them to be as widely available as possible. There are still compute nodes on El Gato with 90 nodes provisioned with Nvidia Tesla K20's.

Ocelote is built on CentOS 6 which is the foundation for all compilers, libraries, and applications.  Some workloads need CentOS 7 for its newer kernel.  A number of GPU nodes have been reprovisioned with a CentOS 7 image. PBS information 

Specifications

Image Removed

. The limitation is a maximum of 10 concurrent jobs. One node with a V100 is also available. Since there is only one, you can feel free to use it for testing and comparisons to the P100, but production work should be run on the P100's. There is also one node with two P100's for testing jobs that use two GPU's. This one should be used to compare with running a job on two nodes.  

Image Added

Cuda Modules

The latest Cuda module available on Ocelote the system is 10.1  Prior versions are still available.
11.0 and is the only version until newer ones come along. The Cuda driver version on each P100 is 418.87
module load cuda will load cuda10.1/toolkit/10.1.168
You may also need module load cuda101/neuralnet in case you need the cuDNN libraries for Tensorflow
module load tensorrt will provide support for tensorflow, particularly the library called libnvinfer.so.6 can be queried with the nvidia-smi command. To see the modules available, in an interactive session simply run:

$ module avail cuda

-------------------- /opt/ohpc/pub/moduledeps/gnu8-openmpi3 --------------------
   cp2k-cuda/7.1.0

-------------------------- /opt/ohpc/pub/modulefiles ---------------------------
   cuda11-dnn/8.0.2    cuda11-sdk/20.7    cuda11/11.0

OpenACC

The OpenACC API is a collection of compiler directives and runtime routines that allow you to specify loops and regions of code in standard C, C++, and Fortran that you can offload from a host CPU to the GPU.

We provide two methods of support for OpenACC

1. We support OpenACC in the PGI Compiler.  The PGI implementation of OpenACC is considered the best implementation.  
   "module load pgi" on Ocelote. If you are on a GPU node from an interactive session you can run "pgaccelinfo" to test functionality.  Remember that the login nodes do not have GPUs or software installed.  
   A useful getting-started guide written by Nvidia is atavailable here:
    https://www.pgroup.com/doc/openacc17_gs.pdf 
   
   
2. We support OpenACC in the GCC Compiler 6.1 which is automatically loaded as a module when you log ininto Ocelote.  Verify with "module list".
   The GCC 6 release includes a much improved implementation of the OpenACC 2.0a specification.
   A useful quick reference guide can be  found at:
   is available from: https://gcc.gnu.org/wiki/OpenACC#Quick_Reference_Guide

About two times a year we host the Xsede Workshop on Programming GPU's with OpenACC.  These courses provide an overview of how to accelerate your code without a lot of programming knowledge.  Watch for announcements to the HPC-Info list.
https://www.psc.edu/xsede-hpc-series-all-workshops 

Applications

Many applications have been optimized to run faster on GPU's.   These include:

NVIDIA GPU Cloud Container Registry

We support the use of HPC and ML/DL containers available on NVIDIA GPU Cloud (NGC). Many of the popular HPC applications including NAMD, LAMMPS and GROMACS containers are optimized for performance and available to run in Singularity on Ocelote. These containers must be run on nodes with CentOS 7 installed.

The containers and respective README files can be found at /unsupported/singularity/nvidia/os7

NGC also provides a set of popular ML and DL frameworks which is not a trivial task. They have made them available to us and they will be updated regularly.  They are currently located at: /unsupported/singularity/nvidia

The other option for accessing ML / DL frameworks is through Python 3.6 which is covered in the next section.

Current list:

PyTorch is a Python package that provides two high-level features:

• Tensor computation (like numpy) with strong GPU acceleration
• Deep Neural Networks built on a tape-based autograd system

nvidia-mxnet.18.09.simg

Each is provided in a Singularity container.

The file name has a tag at the end that represents when it was made, so 18.01 is January 2018

Pulling Nvidia ML / DL Images on Ocelote

Follow this procedure.

Python ML/DL including Nvidia RAPIDS 

The minimum version of Python that is supported is 3.6 so you will module load python/3.6 for all these functions. This will get you to installed packages including:

PBS Usage

Either: copy the file you wish to use to your directory.  Your home path as well as /extra and /xdisk have been bound to the image, so those are your choices.

Or: run the singularity file from where it is.  Since you cannot modify it you will not interfere with anyone else. 

For interactive use, start an interactive job on a GPU node modifying this command:

$ qsub -I -N jobname -W group_list=GROUP-NAME -q windfall -l select=1:ncpus=28:mem=224gb:np100s=1:os7=True -l walltime=1:0:0

You must change the group_list and you should change the other attributes as desired.

On the compute node assigned to you, as an example you can run:

$ module load singularity
$ singularity exec --nv nvidia-tensorflow.18.01-py3.simg python tensorflow_example.py

You need to include the --nv and note it has two dashes.  This will bind the Cuda libraries.
The example file is included in this directory. "tensorflow_example.py"

For batch use, you will include these three lines in your submission script

#PBS -l select=1:ncpus=28:mem=224gb:np100s=1
module load singularity
singularity exec --nv nvidia-tensorflow.18.01-py3.simg python tensorflow_example.py

You will want exclusive access to the node so there is not contention for the GPU.  That is obtained by asking for all 28 cores as shown above

There are more detailed examples here

Singularity

For more information on Singularity, see their web site at:

http://singularity.lbl.gov/user-guide

There are tutorials for Singularity on HPC here

Training

We host workshops from the Pittsburgh Supercomputer Center which is a NSF funded location.  We are working with Nvidia to offer a workshop in the April 2018 timeframe.

Watch for announcements from the hpc-info list.