GPU Programming Competition

Join the Thinkin' Rocks GPU Programming Competition to get hands-on with high-performance computing! A friendly first edition: come to learn, compete, and help us shape future competitions.

The event is open to engineers, students, and anyone who likes making code run blazingly fast. We emphasize understanding how the hardware works and squeezing performance out of it, with no AI tools.

The competition is a 5-hour sprint where you compete on a leaderboard against other contestants for having the fastest solution for 3 difficult parallelism problems.

You should be comfortable writing basic C/C++. No prior GPU or CUDA experience is required, since you’ll have the manuals and documentation to learn from during the event. Bring a laptop to write and submit your solutions.

Make sure to preread the available materials! It's useful to know how to navigate them.

Timeline

10:00 Doors open, set up your environment, test connection.

11:00 Competition starts. Energy drinks are available.

16:00 The leaderboard freezes, solutions are reviewed and everyone enjoys pizza.

Structure

You are provided 3 different problems, each with a baseline solution.

All code runs on a constrained remote machine with an RTX 5070 and a our custom grader.

Each problem has correctness and benchmark tests. Your goal is to score the highest on the benchmark while staying correct.

You are provided a list of useful resources to help optimize the code:

https://cp-algorithms.com/index.html

https://cppreference.com/

https://docs.nvidia.com/cuda/cuda-programming-guide/

https://docs.nvidia.com/cuda/cuda-c-best-practices-guide/

https://0xsero.github.io/blackwell-gpu-wiki/

https://mlc.ai/modern-gpu-programming-for-mlsys/

https://gau-nernst.github.io/tcgen05/

Submissions are accepted in C/C++.

No LLMs allowed.

For any questions, message @ktnlvr on Telegram or email [email protected].