Modular Tech Talks is a behind-the-scenes series featuring internal presentations from our engineering team, offering a deep dive into the Modular technology stack.
In this edition, recorded on March 25, 2025, Weiwei Chen and Fabio Riccardi explore the unique architecture of the Mojo compiler and share how we address the challenges of developing kernels for modern GPUs.
Save our Internal Tech Talks playlist to follow along with future tech talks! https://www.youtube.com/watch?v=Invd_dxC2RU&list=PLh0S94-sJw_6ygGMynvQkt32IwBJM4DBW
00:00 Intro
00:59 Mojo at a glance
02:00 Mojo compilation flow
06:58 Mojo compiler MLIR dialects
10:39 Mojo compilation TLDR
15:57 Mojo dev tools
17:47 The challenge of dense linear algebra
20:34 GPU programming complexity
21:57 Pipelined GPU kernels
23:52 Specialized GPU hardware
25:16 Mojo as a systems programming language
27:13 MLIR: the foundation of hardware abstraction
28:50 Modular's GPU programming model
30:04 Mojo code example
30:44 Mojo's metaprogramming power
32:06 Layout algebra
34:42 Pipeline management
35:09 Performance advantages
36:12 Conclusion
37:36 Q&A
Join our community 🤝:
Forum - https://forum.modular.com/
GitHub - https://github.com/modular
X (aka Twitter) - https://x.com/modular
LinkedIn - https://www.linkedin.com/company/modular-ai
Reddit - https://www.reddit.com/r/ModularAI/
#gpu #programming #ml
