CHAPTER 4: GPU Computing Basics with CUDA C

Author: Libby Shoop, Macalester College

To Cite:

Shoop, Libby. “GPU Computing Basics with CUDA C”. PDC for Beginners, edited by CSinParallel. 2022. Available Online.

Computing using GPUs is often reffered to as massively parallel computing, because we use Graphical Processing Unit devices that contain thousands of cores.

Manycore computing is an exciting technology that drives a great deal of cutting edge computing in Artificial Intelligence (inparticalur Deep Learning in generative AI), Robotics, and a variety of scientific research, such as weather forcasting, biofinformatics, genomics, proteomics, and modeling of various physical and chemical systems.

For NVIDIA GPU cards, one software system to code in is CUDA (Compute Unified Device Architecture), which has compilers for C, C++, and Fortran.

Language used for these examples: C, with special additions for compiling for executing on a GPU, referred to as CUDA C.

• 4.1 The manycore GPU device in a host
  • Getting the code for these examples
  • Discovering Information about your GPU
  • GPU Architecture
• 4.2 GPU Programming model
  • The dim3 data structure and the CUDA programming model
  • Executing code on the GPU device
  • Test the limits
  • Review questions
• 4.3 Mapping Threads to Data Elements
  • 1D grid of 1D blocks of threads
  • Exercises
  • 2D grid of 2D blocks of threads
• 4.4 Example: Vector Addition
  • Three new indispensable additions to the code
  • Test your understanding so far
  • The steps of vector addition
  • Obtaining an array index for a thread
  • Run it here
  • Build and run on your machine
  • Exercises
• 4.5 Vector Addition example with CUDA Unified Memory
  • This method is now preferred
  • The main program
  • Run it here
  • Build and run on your machine
  • An Exercise

You have attempted of activities on this page