Explore Nvidia CUDA

Question

I am a C ++ programmer who develops image and video algorithms, should I learn Nvidia CUDA? or is it one of these technologists that will disappear?

Answer 1

CUDA is currently the only vendor technology from NVIDIA and therefore does not support multi-vendor support, OpenCL does.

However, it is more mature than OpenCL , has excellent documentation , and acquired skills will be easily transferred to other tools for processing parrallel data.

As an example, read Parallel data algorithms from Steele and Hillis, and then look at Nvidia Tutorials - there is a clear connection between them, but Steel / Hillis's article was written over 20 years before the introduction of CUDA.

Finally, FCUDA Projects is working to allow CUDA projects to focus on nvidia hardware (FPGA).

Answer 2

CUDA should be followed for a while, but if you are just starting out, I recommend looking at OpenCL or DirectCompute . Both of them work on ATI, as well as NVidia hardware, and also work on vector units (SSE) of processors.

Answer 3

I think you'd better stick with OpenCL , which is an open standard and supported by ATI, nVidia, and others. CUDA may not disappear in the coming years, but in any case, it is not compatible with GPUs other than nVidia.

Answer 4

OpenCL may take some time to become ubiquitous, but I found the CUDA training very informative, and I don't think CUDA will be out of focus soon. In addition, CUDA is simple enough that the time taken to learn it is much shorter than the CUDA shelf life.

Answer 5

This is the era of high-performance computing, parallel computing. CUDA and OpenCL are the latest GPU Computing technologies that are actually high performance computing! If you are a passionate programmer and want to achieve a benchmark in parallel algorithms, you should really go for these technologies. The parallel part of your program will run for a split second on the multi-core GPU architecture, which usually takes much longer on your processor.