Satori Login

You can connect to Satori through

Web Portal Login

To use the web portal

  1. Navigate to in your favorite web browser.

  2. Select the Globus Auth button.

  3. When redirected to globus, authenticate via Massachusetts Institute of Technology.

  4. When successfully logged in, you will be redirected back to Satori Portal!

you can then launch Jupyter notebook environments running on GPUs

  1. Select “Jupyter Notebook” from the “Interactive Apps” dropdown in the top navbar. You can also select “My Interactive Sessions”.

  2. At the prompt to launch a Jupyter notebook, select the “Anaconda 2019.10 PowerAI [Experimental]” option in the “Choose Anaconda Environment” section.

  3. Enter (in hours) how long you would like the app to run for.

  4. Click launch, wait for app to begin running on a compute node and finally, select “Connect to Jupyter”!

the web portal also has sub-menus for

  1. a terminal on the front-end node (Clusters menu)

  2. a running job monitoring window (Jobs menu). This is also has a batch job preparation form, but it is malfunction at the moment!

  3. a file browsing and transfer client (Files menu)

SSH Login

Satori Login
  • Login to Satori cluster is done via SSHv2 as follow:

  • To copy files from your workstation to Satori you can use SCP, WinSCP, Cyberduck etc

scp <>

As general rules:

  • - should be used for submiting training jobs and related activities

  • - should be used for transfering large files/datasets and compiling software requiring nvcc, gcc, XL compiler etc

  • if one login-node will not be available try the second one

  • don’t run large computations on the login nodes

In case you want to work with Jupyter Notebooks but cannot use the portal, you will need to forward Jupyter HTTP ports to your own workstation via SSHv2 tunneling. This can be done like this: (Note FQDN == Fully Qualified Domain Name like nodeXXXX for the node you are running on (E.g. node0014)

ssh -L 10001:<compute_node_FQDN>:10001
jupyter notebook --ip=<compute_node_FQDN> --port=10001

if you do this, remember to activate whatever conda enviroment you wish to use before starting jupyter e.g

conda activate wmlce-1.6.2