Introduction to GPU computing on HPC: GPU and SSH setup

If you want to follow along on your local machine use these instructions!

If you only want to follow along using Artemis, all you need is an SSH client, as described below.

We will be using the Artemis Training Accounts today (instructions detailed below.)

All the data is on Artemis in /project/Training/GPUtraining (which you can access through the training accounts) or you can download the data from here

GPU Requirements

NVIDIA local installation instructions

Windows 10

Install Visual Studio 2017.

Install the NVIDIA graphics driver and CUDA drivers. Download both from the NVIDIA download page.

Specific versions of tools working together for me are:

C compiler, installed with Visual Studio 2017, cl.exe

Microsoft (R) C/C++ Optimizing Compiler Version 19.16.27027.1 for x64

Nvidia cuda compiler (installed with the CUDA toolkit), nvcc

nvcc: NVIDIA (R) Cuda compiler driver Cuda compilation tools, release 10.0, V10.0.130

You may need to link the correct cl.exe and nvcc somehow. The way I acheieved this was by launching the x64 Native Tools Command Prompt from the Developer command prompt shortcuts listed here and then I could run nvcc successfully.

Linux (Xubuntu 18.04)

Probably as simple as selecting the NVIDIA driver. Then installing the CUDA drivers for the driver/GPU combo. More info here

Specific versions of tools working together for me are:

C compiler, gcc

gcc (Ubuntu 6.5.0-2ubuntu1~18.04) 6.5.0 20181026

and the Nvidia cuda compiler (installed with the CUDA toolkit), nvcc

nvcc release 9.0, V9.0.176


If you have Mac product newer than about 2014 you probably don’t have CUDA-capable GPU card in there. This was done for various reasons. Nevertheless, there are still drivers from NVIDA, and a few options with external GPUs. But good luck, you are on your own. For now, you can do the Artemis examples!


I will be using Matlab 2018a locally and on HPC.


I will be using Python 3.5 - further setup instructions are given in the Python example page.

Which version(s) to use?

Depends what features you need; if you need the latest then go with that. Different cards have different compute capability and different CUDA requirements. And these options bleed down the dependency list (also known as software stack). So if you don’t know, go for the latest stable release compatible over your software stack. Check your CUDA compatability here

There is an update for XXXX, should I get it? Maybe, but probably not. (Not while I am teaching you anyway.) This workshop is for scientific development of applications, chances are you will hack together some code and run it once, so we are not aiming to get this working on every GPU system in the world. But versioning is super important.

What about Containers?

Docker/singularity are great for simplifying the development environments, BUT they still require the underlying installitions of NVIDIA drivers for your specific GPU card, plus a version of CUDA that works with that combo!

For HPC work all you need is an ssh client as below…

Get a shell terminal emulator to connect to Artemis HPC

To connect to Artemis HPC, and follow this lesson, you will need a ‘terminal emulator’ program installed on your computer. Often just called a ‘terminal’, or ‘shell terminal’, ‘shell client’, terminal emulators give you a window with a command line interface through which you can send commands to be executed by your computer.

More precisely, these commands are executed by your shell, which is a program designed to do just that: execute commands! The most commonly used shell is ‘Bash’, and we’ll generally refer to the shell as a ‘Bash shell’, and to scripts as ‘Bash scripts’. There are other shells, and each has its own set of extra commands or syntaxes it can accept – its own scripting language.

The good old days..

You don’t need to worry too much about this! You just need a shell – almost all will understand the commands we’re going to be using.

Linux systems

If you use Linux, then chances are you already know your shell and how to use it. Basically, just open your preferred terminal program and off you go! An X-Window server (X11) may also be useful if you want to be able to use GUIs; again, if you’re using Linux you probably have one, and if you don’t have one, it’s probably because you intentionally disabled it!

OSX (Mac computers and laptops)

Mac operating systems come with a terminal program, called Terminal. Just look for it in your Applications folder, or hit Command-Space and type ‘terminal’. You may find that other, 3rd party terminal programs are more user-friendly and powerful – I use Iterm2.

Terminal is OSX's native terminal emulator.

We also recommend installing XQuartz, which will replace OSX’s native X-Window server. XQuartz has some extra features that may offer better performance when using GUI programs. You’ll need to log out and back in again after installing XQuartz in order for it to activate.


If you’re using a Windows machine, don’t panic! You might not have used ‘CMD’ since Windows 95 but, rest assured, Windows still has a couple of terminal programs and shells buried in the Programs menu.

However, those aren’t going to work for us, as you’ll need extra programs and utilities to connect to Artemis, such as an SSH implementation. To use Artemis on Windows, you have a couple of options:

X-Win32 is full-featured X-server and terminal emulator for Windows. USyd provides a license for it; however, the download link is restricted to staff so, students, get a copy here. Install, and follow the instructions on the USyd-ICT page to activate – you’ll need to be on the USyd network or VPN to do so.

Windows Defender and internet permissions

Windows Defender (Windows’ firewall/antivirus) will probably tell you that certain access has been blocked for X-Win32 and its built-in sound server.

You should allow all connection domains (check all three boxes) in both of these dialogue windows, so that X-Win32 can work on all types of internet connections.

Or, choose the ones you want if you know what you’re doing!

Then setup as follows:

  1. Create a new ‘Manual’ connection, and select ‘SSH’

  2. Fill in the connection details:
    • Name: eg “Artemis”
    • Host:
    • Login: <Unikey>, or Artemis training account ict_hpctrainX
    • Command: /usr/bin/xterm -ls -fa 'consolas' -fs 11
    • Password: <password>

  3. Click ‘Save’


PuTTY, an SSH and telnet client, is another good option if you can’t or don’t want to install X-Win32. However, note that PuTTY does not provide an X11 server, so you won’t be able to use GUI programs on Artemis with just PuTTY.

Head to and download PuTTY. You can install it to your computer, or just download the ‘binary’ and run it directly. Create a new session for use with Artemis as follows:

  1. Fill in the connection details:
    • Host Name:
    • Port: 22
    • Connection type: SSH

  2. Name this session “Artemis” and click ‘Save’

Off-campus access

If you’re attempting this training by yourself, or following on Zoom, off-campus then you’ll need to connect to the USyd internet network before you can connect to Artemis.

There are a couple ways to do this:

1. The USyd VPN

VPN (Virtual Private Network) is a protocol that allows you to tap into a local private network remotely. Follow USyd ICT’s instructions here. Once you’ve connected to the VPN, the above connection methods will work, just as though you were on-campus.

2. Use the Artemis Jump server

Artemis provides a ‘gateway’ server, called Jump, that allows connections from outside the University network, and is itself on the network. From the Jump server, you can then connect to Artemis directly. If using the Jump server, you will need to edit the host address used in the instructions above:

  • Instead of use

This will connect you to Jump, rather than Artemis itself. You can then connect to Artemis directly via SSH. See Episode 1 of the Introduction to Artemis HPC course.

Graphical login nodes

There is one final way to access Artemis, and that is using our graphical login nodes. These special are graphics-enabled login servers which host ‘NoMachine’, a kind of remote desktop service.

To use the graphical login nodes:

  • Download and install the NoMachine Enterprise Client for your operating system. Please don’t download the client from NoMachine directly, as their current version may not work with Artemis.
  • If you’re using MacOS, you need to install XQuartz.
  • Download the glogin.nxs session file to your computer.

To connect:

  • Double-click or run the glogin.nxs shortcut you downloaded (it will open NoMachine Client).
  • A NoMachine window should start, asking for a username and password.
  • Replace “YOUR_UNIKEY” with your UniKey, then enter your UniKey password in the password field, then click ‘OK’.
  • A short while later, an Artemis terminal window should open.