does anyone know the bounds on the size of cudamemcpy(). The size of the result array that I am reading back from the GPU is 2000sizeof(float) but I get all zeros in my result array. when I decrease it to 1000sizeof(float) I get correct answers.
2000 * sizeof(float) is very small amount of memory…I don’t think the problem is related to the size limit, which is very large, but other part of your program.