Open Source Graphics 101: Getting Started
Talkat Open Source Summit Europe + Embedded Linux Conference Europe 2019 · on Tuesday, 2019-10-29· at 15:15· Forum 1·
summary
Quick tour of how GPUs turn models and textures into pixels, claims that workloads are massively parallel (think millions of models and 1080p/4K pixel counts) and that performance hinges on SIMD, many tiny cores, and latency-hiding multithreading. Explains the vertex to fragment pipeline, kernel vs userspace split, command-stream layout and where buffers live, and the open-source stack centered on Mesa with Mesa NIR compiler and GEM. Recommends starting on one feature or bug, shows Gallium vs Vulkan driver structure, and notes practical limits when trying to emulate CPU-style caches.
AI-written summary · about
chapters (10)
- 0:00Introduction to GPUs
- 1:35Basic GPU functions
- 3:31The geometry stage
- 7:06The rasterizer stage
- 9:01GPU hardware architecture
- 15:35Interacting with the GPU
- 17:28Graphics APIs and stack
- 21:41Kernel and user drivers
- 25:01Mesa and driver design
- 30:49Questions and answers
related talks
01h 16mOpensource GPU Drivers BoF
37mProgress in the Embedded GPU Ecosystem
25mOpen Source Model Performance Optimization With SGLang
48mOpensource GPU Drivers BoF
32mImproving the Automotive Experience Through Open Source 3D Engines
20mOpen-Source FPGAs: Tools, Flows, and Visualization
37mHow Igalia is Driving Innovation in Embedded Systems with Open...
20mTrue Exponential: Building a Generative AI Company