How do I use nv-nsight-cu-cli and the GUI version for profiling?

Developer Tools Nsight Compute
1

I was trying to visualize the struct of a graph. The code I’m using is:
https://github.com/NVIDIA/cuda-samples/tree/master/Samples/simpleCudaGraphs
However, whenever I run the command nv-nsight-cu-cli ./simpleCudaGraphs it just says “==PROF== No kernels were profiled”. When I use nvprof it will yield output for kernel execution time. How do I use nsight compute to profile the code? When I use GUI it will throw error “Connection error detected communicating with target application” when I just let it run. The profile option in the GUI is also confusing to use. I can’t select a file to open because it doesn’t exist or it will automatically add the file file extension at the end. I find the official documentation completely useless to solve this problem. How do I use the nsight compute properly? Thanks!

System: Ubuntu 18.04.2 LTS
GPU: 4xTesla P100
nvcc version: 10.1

2

Did you run nv-nsight-cu-cli with root permission? Adding sudo resolved my “No kernels were profiled” situation.

3

Yes. But when I do sudo /usr/local/cuda-10.1/NsightCompute-2019.1/nv-nsight-cu-cli ./app it throws another error saying

==ERROR== The application returned an error code (11)
==WARNING== No kernels were profiled

This only happens when I ssh into my Telsa P100 but runs without problem when I test it on my GTX1080 locally. Any idea why?

4

It seems that your program were aborted. (return code 11 perhaps segmentation fault)
Does it run normally without using profiler on P100?