What do you think?
Rate this book
CUDA C++ Optimization: Coding Faster GPU Kernels
Increase the efficiency of CUDA C++ kernels for AI and high-performance computing on the powerful NVIDIA GPUs. Leverage your GPU investment with the power of an efficient software layer.
Main Topics
- Speeding up CUDA C++ kernels
- Parallelization and vectorization
- Compute optimizations
- Memory access optimizations
Table of
1. Parallel Programming
2. Optimizing CUDA Programs
3. Vectorization
4. AI Kernel Optimization
5. Profiling Tools
6. Compilers and Optimizers
7. Timing CUDA C++ Programs
8. Memory Optimizations
9. Coalescing and Striding
10. Data Transfer Optimizations
11. Heap Memory Allocation
12. Compute Optimizations
13. Warp Divergence
14. Grid Optimizations
15. Compile-Time Optimizations
16. Arithmetic Optimizations
17. Floating-Point Bit Tricks
18. Advanced Techniques
CUDA C++ Slugs
Main Topics
- Speeding up CUDA C++ kernels
- Parallelization and vectorization
- Compute optimizations
- Memory access optimizations
Table of
1. Parallel Programming
2. Optimizing CUDA Programs
3. Vectorization
4. AI Kernel Optimization
5. Profiling Tools
6. Compilers and Optimizers
7. Timing CUDA C++ Programs
8. Memory Optimizations
9. Coalescing and Striding
10. Data Transfer Optimizations
11. Heap Memory Allocation
12. Compute Optimizations
13. Warp Divergence
14. Grid Optimizations
15. Compile-Time Optimizations
16. Arithmetic Optimizations
17. Floating-Point Bit Tricks
18. Advanced Techniques
CUDA C++ Slugs
233 pages, Kindle Edition
Published October 14, 2024
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 of 1 review
February 19, 2026
Good:
clear coverage of the basics
Witty and enjoyable
digresses into general C++ optimization (which is not a bad thing)
emphasizes importance of bit-twiddling hacks performance black magic
A small poke at PTX
Not so good:
lacks some canonical code samples for kernel patterns - eg tiled memory (with padding to avoid bank conflicts) for block matmul
pedantic on error checking macros
not enough on ncu analysis, grid-block sizing
clear coverage of the basics
Witty and enjoyable
digresses into general C++ optimization (which is not a bad thing)
emphasizes importance of bit-twiddling hacks performance black magic
A small poke at PTX
Not so good:
lacks some canonical code samples for kernel patterns - eg tiled memory (with padding to avoid bank conflicts) for block matmul
pedantic on error checking macros
not enough on ncu analysis, grid-block sizing
Displaying 1 of 1 review