Presented at
DEF CON 33 (2025),
Aug. 9, 2025, 2 p.m.
(240 minutes).
There is a creature that lives inside our smartphones, laptops, and PCs, quietly driving their most cutting-edge behaviors. Much larger versions of it hide in datacenters around the world, constantly crunching through massive computation problems. And yet, even experienced engineers find it mysterious. Originally made to boost graphics performance, it has evolved into the engine that powers technologies behind systems like Claude and ChatGPT. In this workshop, we will uncover the nature of this creature: the GPU. Starting with its history and evolution, we will explore how a processor meant to accelerate 3D graphics became the driving force behind modern machine learning and AI. Along the way, we will dive into the design and behavior of neural networks, and discover how a machine built for graphics rendering learned to interpret images and speak human language. Finally, we will investigate how the complexity of neural networks made possible by GPUs can lead to unexpected and strange behaviors... some of which may not be accidental.
Presenters:
-
eigentourist
Eigentourist is a programmer who learned the craft in the early 1980s. He began formal education in computer science when the height of software engineering discipline meant avoiding the use of GOTO statements. Over the course of his career, he has created code of beautiful simplicity and elegance, and of horrific complexity and unpredictability. Sometimes, it's hard to tell which was which. Today, he works on systems integration and engineering in the healthcare industry.
Similar Presentations: