Ncu no kernels profiled -- Target process xxx terminated before first instrumented API call

Developer Tools Nsight Compute
, ,
1

Hi,

I’m trying to use nsight compute to profile a large language model program. And I use ncu --kernel-name my_kernel --launch-skip 86 --launch-count 1 --target-processes all python run.py to obtain the kernel result.

But I always got this error:
==PROF== Target process 112708 terminated before first instrumented API call.
which results in ==WARNING== No kernels were profiled.

What can I do to solve this problem?
Any help would be so appreciated!

2

Hi, @hyaloids

Sorry for the issue you met !
Would you please try a simple CUDA sample to see if NCU works firstly ? This will help us to isolate if this is an ENV issue or specific sample issue.

3

Hi, its a really large project so I need some time turning it into a simple version or even worse may not be able to get the simple version.

4

You can get CUDA samples from GitHub - NVIDIA/cuda-samples: Samples for CUDA Developers which demonstrates features in CUDA Toolkit

5

you can try use ncu 2025.01 version ,
in 2023.01 ,i got same error , strace log recorded some fd close by itself before write it

6

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.