What is the best architecture to use for GPU computing
You need to move the calculations from CPU to GPU. I started Googling about it all, as a result, there are 2 similar architectures - CUDA from Nvidia and OpenCL from AMD, the second can also be used on Nvidia. It seems to be written that the graphics cards on AMD are more focused on computing and faster by 30-40%.
Who had experience programming on such a thing. Can you please clarify what is better to choose for yourself?
I write the program in C#, but as I understand it, the calculation class will have to be written in C++ and connect the library to your program.
1 answers
I have been doing complex hydrodynamic calculations on the GPU for a long time - OpenCL. In terms of computing performance float, NVidia cards are always ahead of AMD with the same number and frequency of stream processors. By 30-50 percent. It is better to avoid using double whenever possible.