Beyond CUDA: GPU Accelerated Machine Learning on Cross-Vendor Graphics Cards with Vulkan Kompute

Explore GPU-accelerated machine learning beyond CUDA using Vulkan Kompute for cross-vendor graphics cards in this conference talk. Delve into parallel processing options and the advantages of Vulkan over traditional C++ SDKs. Learn about Kompute, a general-purpose Vulkan compute framework, and its components. Follow along with a practical machine learning example, including setup, shader logic, tensor creation, and parameter learning. Gain insights into implementing linear regression using Kompute and understand the high-level roadmap for GPU-accelerated machine learning across various hardware vendors like AMD, Qualcomm, and NVIDIA.

Syllabus

Intro
Hello, my name is Alejandro
High level Objectives
Why Parallel Processing?
Parallel Processing: Options
Introducing Vulkan
Vulkan C++ SDK Disadvantages.
Enter Kompute The General Purpose Vulkan Compute Framework.
Vulkan Kompute: Components
The Hello World of ML
ML Example Intuition
Kompute Logic to Set Up
LR Shader Logic
Kompute Logic: Create Tensors
Kompute Logic: Init Tensors
Kompute Logic: Main Sequence
Kompute Logic: "Learn" LR Params
Kompute Logic: Print LR Params
High level Roadmap