Solid Optimization Tips for CUDA

While browsing through “CUDA Fortran for Scientists and Engineers”, I noticed that the authors provide valuable optimization tips. That piqued my curiosity.

I am not a Fortran programmer myself, so I cannot judge the book from that perspective. But as a C++ developer, this book offers many valuable insights into optimizing CUDA kernels. As an example, the mathematically precise analysis of the “Fused Multiply Add (FMA)” operation in Section 5.2.1 can be mentioned. The book was written by two very competent authors and is very solid. Both authors work at NVIDIA and therefore have insider information.

Unfortunately, the book is already a bit dated; the current Maxwell architecture was only released in 2014. The book covers the Kepler architecture up to Compute Capability 3.5.

However, I have to deduct one point because 61 pages of source code were printed in the appendix. That is no longer up to date; you can just download it.

• Gregory Ruetsch, Massimiliano Fatica
• CUDA Fortran for Scientists and Engineers
• Morgan Kaufman
• 2013

See also the review on Amazon.