Home

NVIDIA CUDA Knowledge Base

Welcome to the knowledge base. All you need to get started is a CUDA-enabled device that you can run the example programs on, a basic understanding of C/C++ (some of the more complex elements like memory management and pointers will be reviewed as needed), and the desire to learn.

Fundamental CUDA Concepts

1. What is Parallel Computing?
2. What is a GPU?
3. Basic CUDA Syntax
4. Memory Management on the GPU
   a. CUDA Memory Types
   b. Using CUDA Memory
5. Performance Experiment: On-GPU vs Off-GPU Bandwidth
6. Thread and Block Scheduling
7. Common Parallel Applications
   a. Reduction
   b. Matrix Multiplication

Asynchronous Computing Using CUDA

1. Intro to Asynchronous Computing
2. CUDA Streams
3. Asynchronous Memory Transfers
4. Performance Experiment: Multi-stream Parallelism
5. Basic Synchronization Methods
6. Events and Dependencies
7. Performance Experiment: Event-Based Synchronization vs Explicit Synchronization
8. The Graph Model
9. Creating a CUDA Graph using Stream Capture
10. Performance Experiment: Graphs vs Streams vs Synchronous Kernels
11. Performance Experiment: Increasing the Amount of Graph Nodes
12. CUDA Graph API
13. Synchronization & Dependencies Inside CUDA Graphs
14. Using Host Functions in Graphs & Streams
15. Graph API Node Glossary & Usage Examples