Do not buy this item!
My background is Master level CS, some CUDA experience, but none in a professional capacity.
The best part of these 600 pages of text is their promise: learning how to program with CUDA as an already experienced developer. But that's where the good parts end, sadly.
What stands out first and foremost is the incoherent stream of arbitrary facts and figures related to CUDA programming, nVidia GPUs, parallel programming, and profiling metrics of the author's test programs. It's all over the place! Rarely (if ever) are basic or advanced concepts clearly or concisely explained, the index is insufficient at best (best just Google it or read StackOverflow threads), and what information can be distilled has to be extracted piece by piece by the reader. A term will often be used a fair number of times before the author gets around to any sort of definition or comprehensive description. And when he does, he resorts to car analogies, supermarket queues and whatnot like his audience is completely new to programming ('imagine computer memory like shelf storage, and you're the manager of the warehouse' could very well be a phrase used in this text). The author's idea of explaining complex concepts seems to be to just use a different analogy every time a particular aspect of that concept is relevant to the current subject.
I could go on in some detail about other noticeably poor parts of this item, like its wordiness, lack of vision and structure, quality of its figures and tables ("Descriptions are for suckers!"), endless source code listings in huge font ("Got to get that page count up, or we won't be able to charge $45"), lack of overview, and much, much more, but I'll leave that to others.
In conclusion, it is clear to me that this author is an experienced CUDA developer, but he is unable to rise above the material and explain it to people not already at his level. Analogies make sense to experts, but teach the layman nothing. Timing and throughput results of different hardware need an explanation beyond 'has to do with the such and such'. I highly doubt this 'book' took more that a few months to put together; it seems to be an extended report on CUDA program timings on different hardware started on a boring Friday afternoon, with a lot of cruft added to it by the editor to sell it as a book. I will avoid any Shane Cook items in the future.