cuda-samples
cuda-samples copied to clipboard
Zero copy memory usage performance issues
Recently, I have done some performance optimization work. i used zero copy memory in my work. but i met a strange phenomenon. I have a custom struct array, and i use zero copy memory to store the data. this really saved my time. but when i do the same operation for my another custom struct array(using zero copy memory), The whole program is running slowly. all operations are running slowly, like kernel function, other memCopy operations. the firs custom struct array is a little big, it has 640000 elements. it occupys almost 60Mb memory. is this the reason?