Overview

Welcome to the trial of NVIDIA HPC SDK on NVIDIA LaunchPad! The NVIDIA HPC SDK is a comprehensive suite of compilers, libraries, and tools for developing and deploying HPC applications.

This trial consists of a sandbox environment where you can test your own custom codes on GPUs with the NVIDIA HPC SDK.

To initiate the sandbox environment, choose “System Console” in the navigation panel on the left. An interactive SSH session will be created in the browser. In order to bring in all the components of the HPC SDK, begin by launching an interactive session in the prebuilt NVIDIA HPC SDK container image with:

            

            
docker exec -it hpcsdk /bin/bash
        

Once you enter the container, begin by locating the /share directory. This directory is shared both inside and outside the Docker container environment. Any files placed in /share will persist when the container is exited or shut down.

Let’s try compiling and running a custom code sample. Begin by moving to the /share directory and cloning the CUDA Samples Github repository.

            

            
cd /share
git clone https://github.com/NVIDIA/cuda-samples.git
        

For a simple test, let’s try compiling and running the CUDA deviceQuery sample application. This CUDA application provides us with some detailed technical specifications for any GPUs connected to the test system. First we’ll look for the path of our nvcc CUDA compiler, which is passed to the deviceQuery Makefile with the variable CUDA_PATH.

            

            
which nvcc
>> /opt/nvidia/hpc_sdk/Linux_x86_64/22.11/compilers/bin/nvcc

cd /share/cuda-samples/Samples/1_Utilities/deviceQuery
CUDA_PATH=/opt/nvidia/hpc_sdk/Linux_x86_64/22.11/compilers make
./deviceQuery
        

The application will print technical properties for any connected GPUS before terminating with the message “Result = PASS”.

We’ve demonstrated the ability to download codes to the HPC SDK launchpad system, but what about retrieving application logs and other files? Unfortunately, it is not straightforward to provide outside access into the host system over ssh or ftp. It is recommended to make an outbound connection to transfer files off the host system using the “System Console” link on the left navigation panel of this webpage. For example, SSHing into a public-facing university cluster and using the sftp put command. If this is not possible, there is an alternative method to retrieving files from the host system.

Begin by copying the files of interest to the /data/share directory of the host. Next, click the “Jupyter Notebook” link on the left navigation panel of this webpage. This will open a Jupyter Lab instance. You’ll find the Jupyter file browser on the left side of the webpage. The “share” directory listed in the Jupyter file browser panel is mapped to the /data/share folder on the host system. From the Jupyter file browser panel, simply right click on any files of interest and choose “Download”.