In this section, we will cover the newly introduced CUDA profiler tools, that is, Nsight Systems and Nsight Compute. These profilers support the Volta architecture and onwards GPUs. It is major profiler in the Turing architecture GPU. We will cover the Nsight Systems first, before covering Nsight Compute in the next section.
Nsight Systems (https://developer.nvidia.com/nsight-systems) is a system-wide performance analysis tool that can visualize operations in the timeline and easily find optimization points. In terms of the timeline analysis aspects, Nsight Systems provides system-side utilization information so that we can analyze the bottleneck points. We can get Nsight Systems from the NVIDIA website, but CUDA 10 includes Nsight Systems in the toolkit package by default. All we have to do is make sure it is installed correctly.
...