Back to Numerical Computation
Session 10
Numerical Computation

GPU Computing and CUDA Programming for Numerical Methods

5 hours
Duration
8
Materials
6
Objectives
Session Overview

Comprehensive GPU programming for accelerating numerical computations including CUDA programming, memory optimization, and parallel algorithm design for GPUs.

Learning Objectives
By the end of this session, you should be able to:
  • Master CUDA programming model and GPU architecture understanding
  • Implement efficient matrix operations on GPU with shared memory optimization
  • Apply GPU acceleration to iterative methods and eigenvalue computations
  • Optimize memory coalescing and bandwidth utilization
  • Use CUDA libraries (cuBLAS, cuSOLVER, cuFFT) for numerical computing
  • Design hybrid CPU-GPU algorithms for large-scale problems
Course Materials
Download materials for offline study and reference
GPU Architecture and CUDA Programming Guide (70 pages)
Available material
Matrix Operation GPU Implementations
Available material
Iterative Method GPU Acceleration
Available material
Memory Optimization Techniques and Patterns
Available material
CUDA Library Integration and Usage
Available material
Hybrid CPU-GPU Algorithm Development
Available material
Performance Analysis and Profiling Tools
Available material
Complete GPU-Accelerated Numerical Package
Available material