Modular GPU Kernel Hackathon — Now Officially Powered by AMD 🚀
Please note that you'll need to apply via AGI House here to attend this event.
Development of AI/GPU compute kernels has long been constrained by legacy tools: C++ with its memory headaches, Python with its performance ceilings, and platforms that tie developers to a single vendor. Mojo 🔥 is here to change that, offering a modern programming language purpose-built for high-performance, portable computing.
We’re proud to announce that this hackathon is now an official joint event with AMD. All participants will be developing exclusively on AMD Instinct™ MI300X GPUs: next-generation GPUs purpose-built for AI workloads. This is your chance to write kernels for one of the most advanced AI accelerators on the planet, work side-by-side with the Mojo team, and learn from some of the brightest minds in AI.
Whether you're building machine learning primitives, optimizing data movement, or rethinking traditional algorithms from the ground up, you’ll have access to the hardware, language, and support to make it happen—all in one day.
Brought to you by Modular, in partnership with AMD, Crusoe, and GPU Mode, and hosted at AGI House.
📅 Event Schedule (PT)
10:00 AM – Doors open, early-bird networking (meet future teammates!)
11:00 AM – Talks from Chris Lattner, Ramine Roane, Dylan Patel, Simon Boehm, Sasha Krassovsky, Brad Larson, and Jack Clayton
12:00 PM – Hacker project pitches
12:15 PM – Hacking begins + lunch
4:00 PM – Midpoint check-in
6:00 PM – Dinner
8:00 PM – Final demos + presentations
🏆 Prizes
We’re excited to offer first, second, and third place prizes in partnership with AMD:
1st Prize: 500 compute hours on AMD Instinct™ MI300X GPU
2nd Prize: 300 compute hours on AMD Instinct™ MI300X GPU
3rd Prize: 200 compute hours on AMD Instinct™ MI300X GPU
🤝 Why Mojo + AMD MI300X?
This hackathon is your chance to experiment with a high-performance language on world-class AI hardware. Mojo offers composability, portability, and performance, without vendor lock-in. Combined with the raw power of AMD MI300X, this is a rare opportunity to build next-gen kernels on an open and future-facing stack.
🧠 Project Ideas
Implement a transformer attention block from scratch
Port GPTQ to the MI300X
Build a GPU-accelerated FFT
Optimize cumulative sum (cumsum) operations
Design fast image-processing kernels (e.g. NMS, convolution)
Create a blazing-fast Radix sort
🧑💻 Rules
All submissions must include code in a public GitHub repository.
Max team size: 4 (teams may form onsite).
Attendance is limited to approved registrants only.
Redefine GPU programming on cutting-edge AMD AI hardware, with Mojo at the core. What will you build?