Thanks for your greate contributions. I refered model_usage for C++ , and use opengl to display the cuda data from GPU ram sucessfully, but found out that if i don't copy out the output data from GPU, then the next image that uploading to GPU will cost a long time(about 20 to 50ms, the first image upload to GPU only need about 6ms), if I do copy output from GPU to CPU, the copy out cost 20 to 50ms。Any suggestion how to lower the upload time from RAM to GPU?
Thanks for your greate contributions. I refered model_usage for C++ , and use opengl to display the cuda data from GPU ram sucessfully, but found out that if i don't copy out the output data from GPU, then the next image that uploading to GPU will cost a long time(about 20 to 50ms, the first image upload to GPU only need about 6ms), if I do copy output from GPU to CPU, the copy out cost 20 to 50ms。Any suggestion how to lower the upload time from RAM to GPU?