Welcome

Welcome to the MEAP edition of GPU Programming with Triton, and thank you for being here this early.

This book is for machine learning engineers, data scientists, software engineers, and researchers who work with GPUs every day but have never had the chance to understand what is happening underneath. If you write Python comfortably and build machine learning systems, yet words like warps, shared memory, occupancy, and kernel fusion still feel vague, you are who I had in mind while writing it.

My own start with GPUs probably looks familiar. I could build models and use PyTorch well enough, but the GPU itself stayed a black box. I ran code on it without really knowing why one operation was fast and another was slow, or how the kernels behind modern AI systems got written in the first place.

When I went looking for answers, I kept hitting the same two extremes. Some resources were deeply academic and assumed you already knew computer architecture and low-level systems programming. Others handed you working code without ever explaining what the hardware was doing with it. What I actually wanted was something that began from first principles and slowly connected the hardware, the programming model, and the real workloads people run in machine learning. That gap is what this book tries to fill.