We have a rather complicated HPC system where GPUs are at leaf nodes. We want to visualize kernel execution as the system runs in real time. The system is composed of many GPU processes, and we want the visualization to include all GPU processes. As we want use nsys as our system runs in real time as a whole, it is difficult to use nsys on the parent process and to trace all its children. It seems that our only option is to attach to separate GPU processes and get timelines for each of them. Is there a way to combine all files together into one for better view?
So there area couple of options here.
One is that you can load multiple files into the GUI for visualization.
The other is that you can also do multi report statistical analysis, using our new recipe system, see User Guide :: Nsight Systems Documentation