Overview
Explore GPU accelerated computing and optimizations for cross-vendor graphics cards using Vulkan and Kompute in this CppCon conference talk. Gain conceptual and practical insights into the cross-vendor GPU compute ecosystem and learn how to add GPU acceleration to existing C++ applications. Discover how to write a simple GPU-accelerated machine learning algorithm from scratch that can run on virtually any GPU. Understand the projects enabling acceleration across cross-vendor GPUs and how to harness GPU power using the Kompute framework with minimal C++ code. Delve into advanced optimizations leveraging hardware capabilities of graphics cards, including concurrency-enabled GPU queues for significant performance improvements. Cover GPU computing terminology, data parallelism principles, and hardware concepts like GPU queues and queue families. Learn about advancements in new graphics card architectures supporting multiple parallel queue processing workloads for even greater speedups.
Syllabus
Introduction
Objectives
Why Parallel Processing
Intuition
CPU vs GPU Memory
Grids Blocks Threads
Leveraging Heterogeneity
Vulkan SDK
Vulkan Advantages
Complexity Reduction
Data
Pipelines
Sequential Program
Compute Framework
Compute Components
Compute Manager
Explicit Queues
Taught by
CppCon