Hands-On Introduction to HPC and AI¶

SC26 Student Programming Tutorial

Sunday, November 15, 2026

8:30 AM to 4:30 PM CST

McCormick Place, Chicago, IL

What You’ll Learn¶

This full-day, hands-on tutorial takes you from “What is a supercomputer?” to writing parallel programs that run on GPUs and training AI models – all on a real HPC cluster.

Module	Topic	Key Skills
1	HPC Foundations	Cluster architecture, environment modules
2	Job Scheduling with Slurm	Batch scripts, `sbatch`, `squeue`, `scancel`
3	Shared-Memory Parallelism	OpenMP directives, threading
4	Distributed-Memory Parallelism	MPI communication, multi-process programs
5	GPU Programming	HIP kernels, GPU memory management
6	AI on HPC	Inference, fine-tuning with LoRA
7	AI Agents	Building a tool-using agent, capstone challenge

Prerequisites¶

Basic familiarity with a Linux command line (navigating directories, editing files)
Some programming experience in any language (C, Python, etc.)
No HPC or parallel programming experience required

Getting Started¶

1. Connect to the Cluster¶

You will receive login credentials for the AMD University Program (AUP) AI & HPC Cluster. Connect via SSH:

ssh <your-username>@hpcfund.amd.com

2. Clone This Repository¶

cd ~
git clone https://github.com/Students-SC/sc-student-programming-tutorial.git
cd sc-student-programming-tutorial

3. Verify Your Environment¶

Run the environment check script to confirm everything is set up:

bash setup/check_environment.sh

You should see [PASS] for all items. If anything fails, ask an instructor for help.

4. Set Up Your Python Environment¶

Create the Python virtual environment now, before starting the modules. You will use this same environment several times throughout the day, including for the tutorial-provided AI agent and the afternoon AI exercises.

From the repository root:

sbatch setup/setup_venv.sh

This runs as a Slurm batch job, so it may take a few minutes. You will learn what sbatch is doing in Module 2; for now, submit the job and wait for it to finish:

squeue -u $USER
tail -f setup_venv_<JOBID>.out      # Watch the log (Ctrl+C to stop)

When the job completes, verify the environment:

source "$WORK/sc26_venv/bin/activate"
python3 -c "import torch; print(f'PyTorch {torch.__version__} installed')"
deactivate

You only need to create this venv once. Module 2 includes a checkpoint that explains the Slurm command you used here and helps anyone who skipped this step catch up.

Getting Help¶

If you get stuck, ask for help early. Instructors will be in the room throughout the tutorial and can help with cluster access, Slurm jobs, code errors, and conceptual questions.

We may also provide an online help channel, such as Slack, for questions during or after the session. Details will be added here before the tutorial.

Compute (virtual) Node Details¶

In this tutorial, a virtual compute node is a slice of a larger physical compute node. You can think of it like having your own assigned portion of a shared machine: each virtual node gets a fraction of the physical node’s CPU cores, memory, and GPU resources, while Slurm keeps different users’ jobs separated from each other.

Resource	Details
CPU	AMD EPYC (16 cores per virtual node)
GPU	1x AMD Instinct MI210 (64 GB HBM2e) per virtual node
RAM	64 GB system DRAM per virtual node
Partition	`mi2101x`
Scheduler	Slurm
Compiler	GCC 12.2, `hipcc` (ROCm 7.2)
MPI	OpenMPI 4.1.8

Important

Do not run compute workloads on the login node. All programs must be submitted through Slurm (sbatch or srun). This is both good practice and necessary so that all 80+ participants can share the cluster fairly alongside the production research and education workloads being run on the system.

Using AI Assistants¶

You are encouraged to use AI coding assistants throughout the tutorial – just as you would in modern development practice. The ground rule:

Use the agent, but own the answer.

Let the AI help you write code, debug errors, and explain concepts – but make sure you understand what the code does before moving on. Exercises are designed so you need to verify and reason about the AI’s output.

The tutorial agent: aider + a self-hosted coding model¶

We provide a CLI coding agent for you to use throughout the day – no accounts, no API keys, no cost. It’s aider (a popular open-source coding agent) pointed at Qwen3-Coder-30B-A3B-Instruct, served on a dedicated MI300X compute node by your instructors. Exercises that say “ask your AI agent…” refer to this tool.

Launch it from inside the repo after completing the setup steps above:

# cd into a directory you want aider to work in (your code, exercises, etc.)
cd module-03-openmp/exercises

# Launch aider (this activates the venv from Step 4 and points at the local model)
bash ~/sc-student-programming-tutorial/setup/launch_aider.sh

Inside aider, type instructions in plain English. Aider will read and edit files in the current directory. Type /help to see commands, /exit to quit.

Note

If you’d rather use your own coding assistant (Cursor, GitHub Copilot, ChatGPT, Claude, etc.), feel free – the exercises don’t require ours. But the tutorial-provided one works for everyone with no setup beyond what’s already in this README.

How the agent backend works (preview of Module 7)¶

The agent is the same kind of system you’ll build yourself in Module 7: an LLM running on the cluster, served via an OpenAI-compatible API, queried in a reasoning loop. By the end of the day you’ll understand exactly how it works under the hood.

Repository Layout¶

setup/                              Environment check & Python venv setup
module-01-hpc-foundations/          Cluster architecture & environment
module-02-slurm/                    Batch scheduling with Slurm
module-03-openmp/                   Shared-memory parallelism
module-04-mpi/                      Distributed-memory parallelism
module-05-hip/                      GPU programming with HIP
module-06-ai-inference-finetuning/  AI inference & fine-tuning
module-07-ai-agents/                AI agents & capstone challenge

Each module directory contains a README.md with instructions and an exercises/ folder with starter code and templates.

Schedule¶

Time	Session
8:30 to 8:50 CST	Welcome & AI Assistants Intro
8:50 to 9:20 CST	Module 1: HPC Foundations
9:20 to 10:05 CST	Module 2: Slurm
10:05 to 10:20 CST	Break
10:20 to 11:10 CST	Module 3: OpenMP
11:10 to 12:00 CST	Module 4: MPI
12:00 to 12:45 CST	Working Lunch
12:45 to 1:45 CST	Module 5: HIP GPU Programming
1:45 to 2:00 CST	Break
2:00 to 3:00 CST	Module 6: AI Inference & Fine-Tuning
3:00 to 3:50 CST	Module 7: AI Agents & Capstone
3:50 to 4:10 CST	Wrap-Up