I have some questions about NVVP (NVIDIA VISUAL PROFILER).
Question 1: How can I profile kernels from different CUDA processes? For example, can I profile the processes ./cuda1 & ./cuda2 ?
Question 2: I am very intertsed in extracting (at any time) the information that give me how busy is the GPU from the profiling application. At first I though that the metric I need is the number of SMs a kernel is working on. By the way, I do not find this metric in NVVP or NVPROF. Could you please show me how could I achieve it? Additionaly I found out that metrics like achieved_occupancy is helpful as well. Could you please, provide me the full list of metrics that show how my application makes my gpu busy?
Thank you very much in advance!!