March 9, 2026 · NHR@FAU & NHR@TUD
Introduction to CUDA C/C++
Trainee
Completed targeted training on GPU programming using CUDA C/C++, covering memory management, performance modeling, and parallelization on Nvidia accelerators.
I participated in the Introduction to CUDA C/C++ online training course, co-organized by NHR@FAU and NHR@TUD. Instructed by experts from both high-performance computing centers, this intensive session focused on leveraging Nvidia accelerators for scientific computing.
The course provided hands-on experience and theoretical foundations across the following key areas of GPU programming:
GPU Architecture & Parallelism
- Hardware Fundamentals: Basics of GPU architecture and execution models.
- Parallel Design: Strategies for identifying and effectively expressing parallelism within applications.
CUDA Programming & Memory Management
- Core CUDA C/C++: Writing and executing CUDA code specifically for Nvidia accelerators.
- Memory Allocation: Managing host and device memory, and efficiently transferring data between them.
- Advanced Memory: Implementing managed memory and utilizing asynchronous prefetching to optimize data pipelines.
Performance Optimization & Advanced Operations
- Profiling & Analysis: Utilizing command-line and visual profilers alongside lightweight performance models to assess and overcome performance limitations.
- Efficient Reductions: Applying atomic operations, shared memory, and the CUB library for highly efficient reduction operations.