Introduction / Getting Started

“Research computing” is a broad topic, but usually means running specialized software and operating systems not designed for consumers, then analyzing data produced by that software. Most of the time, the operating system in question in Linux, and the software is run by typing how you want to run that program into a command-line interface. These can be intimidating to beginners! We collect our favorite beginner-friendly resources below, and are happy to meet and provide individual support - please Ask CIT for this consultation.

• Prefer video? Check out our recommended Linkedin Learning collection for research computing. Please see our Linkedin Learning page for more info on that service.

• Prefer text? Try out Cornell’s Virtual Workshop Roadmaps, especially their Introduction to Linux.

Lambda Workstations

We have 2 shared-access Lambda workstations with powerful GPUs:

• deeplearning1.geneseo.edu (Ubuntu Linux 20.04 LTS) / 137.238.8.47

• 🚧 deeplearning2.geneseo.edu (Windows 10) / 137.238.8.48 🚧 Under Maintenance Fall 2024

Access

Access is controlled via the Lambda-User AD group - please Ask CIT to be added or otherwise request changes to this group.

You must either be on campus or connected to VPN to connect to these workstations.

Deeplearning1 Remote Access

Web-Based

🚧 EXPERIMENTAL 🚧

Safari will not work 😔

1. Browse to https://deeplearning1.geneseo.edu/remote/connect.html

2. Host: deeplearning1.geneseo.edu

3. Port: 443

4. Username: Your Geneseo short username

5. Password: Your Geneseo network password

6. Leave Connect selected, and don't worry if a different user is selected. Your username and password will be used to route you to your own backend process.

7. Click Connect

Usage notes:

• Click on the 3 horizontal lines (Hamburger) in the top left to launch programs - the most common programs have been added under (Hamburger) > Start > Geneseo

• To leave your session running for you to connect to later, close your browser tab or click (Hamburger) > Disconnect

• To end your session, click (Hamburger) > Server > Shutdown Server - this does not shut down the entire server, and only ends your session.

• Dragging and dropping a file from your local computer onto a connected session will upload it to your Downloads folder in your home directory if it exists, otherwise your home directory. The system will attempt to open uploaded files that it has a registered program for (PDFs and images should work)

  • You could also do (Hamburger) > Server > Upload file

• Running xdg-open /path/to/some/remote/file in a remote terminal will download that file to your local computer through your browser.

• Terminal, Files, Firefox, and many other apps take a few moments to launch in a fresh session - please be patient.

  • Xterm (under System Tools) launches instantly, but Terminal (Gnome) supports tabs, search, copy+paste, and looks nicer.

Command Line / SSH

• If you're new to the command line or SSH, we've found the following LinkedIn Learning courses to be helpful: Command Line and Remote Access.

• The deeplearning1 server is running an SSH server on TCP port 22

• When you first connect via ssh, you will be asked to confirm the host ssh key, which should be one of the following:

  • SHA256:laNR7b1nLDAkCxFF0JMpzGyTCsPPQbjj+TOb9HuSfNQ deeplearning1.geneseo.edu (RSA)

  • SHA256:uVuGpGGB9U5+OEBpKLfatIkUoCf09TjxycK+Ksg6evg deeplearning1.geneseo.edu (ECDSA)

  • SHA256:1C8/YeaK61hFXIfjzjUW4hlJBche1nGs/C19/4U081g deeplearning1.geneseo.edu (ED25519)

Deeplearning2 Remote Access

Remote Desktop Protocol (RDP) on TCP port 3389

Windows has a built-in Remote Desktop Client; on macOS we recommend the official Microsoft Remote Desktop app from the Mac App Store.

The version of Windows on deeplearning2 limits users to 1 Remote Desktop Client user at a time - please log all the way out when you are done (not disconnect, which leaves your session running), so that others may use this machine remotely.

Software

• deeplearning1 (Linux)

  • Miniconda (see section below)

  • Mathematica 13

  • Matlab R2023a in /usr/local/MATLAB

  • Python 3.8

  • Lambda Stack (collection of GPU-accelerated software packages)

    • Machine learning frameworks: TensorFlow, Keras, PyTorch, Caffe, Caffe 2

    • Included GPU software: CUDA, cuDNN

  • IRAF - "Image Reduction and Analysis Facility, a general purpose software system for the reduction and analysis of astronomical data"

  • 3DSlicer 5

  • FSL

  • R 4.3.x

Miniconda

Miniconda (a slightly smaller version of Anaconda) lets you create, populate, and use isolated environments for Linux software. (If you’re familiar with Python virtualenv, this is very similar.)

First, load the conda command into your shell with

source "/deeplearning/miniconda3/etc/profile.d/conda.sh"

After sourcing that file, you can run conda and see all the sub-commands available. Most commonly you will be creating or activating a new environment.

List all conda envs with

conda env list

To activate a conda env:

conda activate name-or-full-path-of-environment

To create a new conda env with a specific version of python (3.9) installed:

conda create -n env-name-you-want python=3.9

There are many additional flags for conda environment creation, and you may want to install a different version of python - see the conda create docs.

Hardware

These workstations each have:

• 4 NVIDIA Corporation GV100 [TITAN V] (rev a1) video cards, each with 12 GB of video memory and 3584 CUDA cores

• 24-core Intel(R) Xeon(R) CPU E5-2650 v4 @ 2.20GHz

• 128 GB RAM

• 1TB SSD

• 3 x 4TB SSDs as ZFS Raidz1, mounted at /deeplearning