Learning CUDA - Personal Notes

Following the content of the book Cheng, J., Grossman, M. and McKercher, T., 2014. Professional Cuda C Programming. John Wiley & Sons..

Before starting:

• The book is based on Fermi and Kepler microarchitectures. Other architectures may not share all features mentioned in the book.
• Not all contents in this book are concise!

Contents of the notes:

• Chapter 2 - CUDA programming model
• Chapter 3 - CUDA execution model
• Chapter 4 - Global memory
• Chapter 5 - Share memory and constant memory

Brief overview of keypoints:

1. Nvidia microarchitectures:

• 2007 - Tesla
• 2009 - Fermi
• 2013 - Kepler
• 2014 - Maxwell
• 2016 - Pascal
• 2017 - Volta
• 2018 - Turing

2. Maximizing Performance
   CUDA programming guide - Performance Guidelines

• From the view of instruction execution: the key is to hide latency by exposing more parallelism
  • Chapter 2 - CUDA programming model
• From the view of memory access: maximize bandwidth utilization
  • Aligned and coalesced memory accesses that reduce wasted bandwidth - Chapter 4, sec 3.1-3.4
  • [unconcise] Sufficient concurrent memory operations to hide memory latency - Chapter 4, sec 3.5