Thread history

Fragment of a discussion from Talk:Jetson/Tutorials/CUDA
Viewing a history listing
Jump to: navigation, search
Time User Activity Comment
No results

These are general CUDA topics, so I recommend that you learn CUDA if you are genuinely interested, such as by reading one of the popular CUDA books or taking a free online course mentioned at "https://developer.nvidia.com/cuda-training". To quickly summarize, the answer is that CUDA takes care of the parallelization for you, so that you just write your CUDA code to be highly parallel, and the CUDA architecture + driver + hardware automatically takes care of using 192 CUDA cores if it is on a Jetson TK1 or 96 CUDA cores if it is on a laptop GPU or 10,000 CUDA cores if it is on a server GPU, etc.

PS: After you learn more about CUDA, you can run the "deviceQuery" sample program on Jetson TK1 to see various info about the GPU such as how many CUDA cores and blocks it has.

23:36, 28 July 2014