GPU & CUDA

I just finished a book on GPU with CUDA, which is composed by several Chinese authors – I mean, it still suffers from two major common drawbacks of most Chinese technical books:

1. listing knowledges directly without any background introduction, which makes the book be more like a manual/handbook, instead of a textbook.

2. pasting the src code page by page, and I think it is totally unnecessary.

Anyway, it is a good book. It’s the first book about GPU/CUDA in China.

Generally speaking, GPU is a computing specialized architecture product: more paralleled computing units and several layers memory layout.  CUDA is a quite simple language extension. When coding using CUDA, you need a clear chart of GPU architecture in your mind. CUDA still exposes some hardware details. In a word, the essence of GPU/CUDA is parallel programming.

It is a big pity that CUDA only works on nVidia video cards until now. It is said more and more video cards will support CUDA. Great, but before it comes true, OpenCL is necessary(it is said there are too little resources for learning)