Glossary
A quick reference for all technical terms used in this guide, listed by chapter.
Chapter 1: Welcome
| Term | Definition |
|---|---|
| supercomputer | A very powerful computing system made up of thousands of interconnected computers (nodes) working together. |
| Command Line | A text-based interface where you type commands to interact with a computer, instead of clicking with a mouse. |
| HPC | High-Performance Computing. The use of supercomputers and parallel processing to solve large computational problems. |
| High-Performance Computing | The use of supercomputers and parallel processing to solve large computational problems. |
| Terminal | The application on your computer that provides access to the Command Line (Terminal on Mac/Linux, PowerShell on Windows). |
| LLM | Large Language Model. An AI model with billions of parameters trained on massive text datasets to understand and generate language. |
Chapter 2: The keys to the castle
| Term | Definition |
|---|---|
| OS | Operating System. The core software that manages a computer's hardware and runs your programs, such as Windows, macOS, or Linux. |
| SSH | Secure Shell. A protocol for securely connecting to a remote computer over the internet. |
| phishing | A type of cyberattack where attackers deceive people into revealing sensitive information, such as passwords, by pretending to be a trustworthy entity. |
| cryptographic | A secure way of protecting information using mathematical algorithms so that only the right people can access it. |
| SSH Keys | A pair of cryptographic keys (one public, one private) used to authenticate your identity when connecting to LUMI. |
| SSH Key Pair | A pair of cryptographic keys (one public, one private) used to authenticate your identity when connecting to LUMI. |
| Private Key | The secret half of your SSH Key Pair. It stays on your computer and should never be shared. |
| Public Key | The shareable half of your SSH Key Pair. You upload it to the LUMI portal so LUMI can verify your identity. |
| passphrase | A password that protects your Private Key, adding a second layer of security. |
| string | A sequence of characters, such as letters, numbers, and symbols, used in computing to represent text. |
| Key fingerprint | A unique short sequence of characters used to identify a longer public key, allowing you to verify that you are connecting to the genuine server and not an impostor. |
Chapter 3: The Command Line
| Term | Definition |
|---|---|
| script | A list of commands saved in a file that LUMI can execute automatically, even while you're away. |
| Shell | The program that interprets and executes your typed commands. |
| CLI | Command Line Interface. Another name for the Command Line - a text-based way to interact with a computer. |
| Command Line Interface | Another name for the Command Line - a text-based way to interact with a computer. |
| directory | The command-line term for "folder." |
| Working Directory | The directory you are currently "inside" in the Command Line. Shown by the pwd command. |
$HOME | Your personal user directory on LUMI, located at /users/<username>. Where you land after logging in. |
| Parent Directory | The directory one level "above" the current one. Reached with cd .. |
Chapter 4: Navigating LUMI
| Term | Definition |
|---|---|
| Node | An individual computer within the LUMI supercomputer. |
| Login Node | A shared "lobby" node where you manage files and submit Jobs. Not for heavy computation. Named uan01, uan02, etc. |
| Compute Node | A powerful node reserved exclusively for running your Jobs. Named like nid007628. |
| Job | Any task you ask the supercomputer to run - a script that trains a model, processes data, or runs a simulation. |
| processor | The "brain" of the computer that carries out instructions and calculations. Often used interchangeably with CPU. |
| CPU | Central Processing Unit. The general-purpose processor of a computer. Handles complex tasks sequentially. |
| GPU | Graphics Processing Unit. A specialised processor that is incredibly fast at doing many simple calculations at the same time. Essential for AI. |
| hardware | The physical components of a computer system, such as processors, memory, and storage drives. |
| Partition | A group of nodes with specific hardware and resource limits (e.g., small-g, dev-g). |
| Slurm Partition | A subdivision within a hardware Partition that sets limits on Job size, duration, and number of concurrent Jobs (e.g., standard, small-g, dev-g). |
| TB | Terabyte. A unit of digital data storage equal to 1,024 Gigabytes (GB). |
/project | Shared project storage (50 GB). Good for files the whole team needs. |
/scratch | High-capacity temporary storage (50 TB). Your main working area for datasets, outputs, and checkpoints. |
/flash | Ultra-fast temporary storage (2 TB). 3x more expensive - use only when your Job needs very fast I/O. |
| LUMI-C | The CPU Partition of LUMI - nodes with powerful processors but no GPUs. |
| LUMI-G | The GPU Partition of LUMI - nodes equipped with AMD MI250X GPUs for AI workloads. |
| walltime | The maximum duration a Job is allowed to run before it is stopped. |
| I/O (Input/Output) | The speed of reading from and writing to a storage drive. |
| NVMe | Non-Volatile Memory Express. A type of extremely fast flash storage drive. |
| GCD | Graphics Compute Die. One independent half of an AMD MI250X GPU chip. Each MI250X contains 2 GCDs, each with its own 64 GB of VRAM. Slurm treats each GCD as a separate GPU. |
| encryption | A method of protecting data by converting it into a secret format. Only people with the correct password or key can unlock and read it. |
| GDPR | General Data Protection Regulation. A European Union regulation concerning data protection and privacy in the EU and the EEA. |
| Data Controller | Under GDPR, the entity that determines the purposes and means of processing personal data. |
| Data Processor | Under GDPR, the entity that processes personal data on behalf of the Data Controller. |
| Data Processing Agreement (DPA) | A legally binding contract between a Data Controller and a Data Processor regulating the secure handling of personal data. |
Chapter 5: AI software on LUMI
| Term | Definition |
|---|---|
| Lustre | LUMI's high-performance filesystem, optimised for large files but sensitive to many small files. |
| pip | Python's standard package manager - a tool that automatically downloads and installs Python libraries from the internet. |
pip list | A command to display all Python packages (libraries) currently installed in your environment or Container. |
| library | A collection of pre-written code that you can plug into your own programs to perform common tasks (e.g., PyTorch for AI calculations). |
| software | The programs, applications, and scripts that run on the hardware to perform tasks. |
| Container | A single .sif file that packages all your software, libraries, and dependencies into one "box." |
| Apptainer | The Container tool used on LUMI. A secure, supercomputer-friendly alternative to Docker. |
| Singularity | The Container tool used on LUMI. A secure, supercomputer-friendly alternative to Docker. |
| Bindings | Configuration that allows a Container to communicate with LUMI's hardware (GPUs, high-speed network). Set up by loading the lumi-aif-singularity-bindings Module. |
| Runscript | A set of default instructions baked into a Container that define what happens when you singularity run it. |
venv | Virtual Environment. A private, self-contained folder where you install Python libraries for a specific project so they do not interfere with other projects on the system. |
| Virtual Environment | A private, self-contained folder where you install Python libraries for a specific project so they do not interfere with other projects on the system. |
| Module | A pre-installed software package on LUMI that you activate with module load. |
| PyTorch | A popular open-source machine learning framework used for deep learning and artificial intelligence. |
| Docker | A popular platform for building and running software in isolated Containers. |
Chapter 6: GitHub
| Term | Definition |
|---|---|
| GitHub | The world's largest platform for hosting and sharing code. |
| Git | The version-control tool that tracks changes to code. Pre-installed for every user on LUMI. |
| Repository | Aka repo. A project folder on GitHub that contains code, configuration files, and history of all changes. |
git clone | Downloads a Repository from GitHub to your LUMI storage. |
git pull | Updates your local copy of a Repository with the latest changes from GitHub. |
Chapter 7: Slurm
| Term | Definition |
|---|---|
| Slurm | The Job scheduler that manages who gets to use which Compute Nodes and when. |
| Batch Script | A text file containing Slurm options and commands that define what resources you need and what to run. |
sbatch | The command to submit a Batch Script to Slurm. |
squeue | The command to check the status of your queued or running Jobs. |
scancel | The command to cancel a running or pending Job. Use scancel <job_id> or scancel --me to cancel all your Jobs. |
| Interactive Job | A Job where you get a live Terminal session on a Compute Node for real-time work (debugging, testing). |
| Billing Unit (BU) | The unit of currency for LUMI compute time. Different resources (CPU, GPU, storage) cost different amounts of BUs. |
srun | The command to start an Interactive Job or to launch a task within an allocated Job. |
| GPU-hours | One of the two currencies of Billing Units on LUMI. Used for compute time on the LUMI-G GPU Partition. |
| CPU-hours | One of the two currencies of Billing Units on LUMI. Used for compute time on the LUMI-C CPU Partition. |
| Environment Variable | A background setting you define before running a program, which tells the program how it should behave (such as telling it which AI model to load or where to save files). |
| vLLM | A fast software engine used to run LLMs efficiently. It is the recommended way to use large language models on LUMI. |
| log | A text file that automatically records the output, progress, and errors of your programs as they run. Slurm saves these as .out files. |
Chapter 8: Graduation & next steps
| Term | Definition |
|---|---|
| MCP | Model Context Protocol. An open standard that lets AI assistants connect to external tools and data sources - for example, letting your chatbot search the LUMI documentation. |