Nsys crashes on first memory access

Developer Tools Nsight Systems Profiling Linux Targets
1

using nsys to profile my current app crashes on the first memory access (tested with cudaMemset and cudaMemcpy/cuMemcpyHtoD). I’m calling it simply nsys nvprof my_app.

Here’s the stack trace I’m getting when compiled in debug mode:

#13   Object "MY_LIB.so", at 0x71217c8a875e, in initFunction(Arg1&, Arg2&, CUstream_st*)
#12   Object "/lib/x86_64-linux-gnu/libcuda.so.1", at 0x71218dee9c89, in 
#11   Object "/lib/x86_64-linux-gnu/libcuda.so.1", at 0x71218dd3ab81, in 
#10   Object "/lib/x86_64-linux-gnu/libcuda.so.1", at 0x71218e0bc99f, in 
#9    Object "/lib/x86_64-linux-gnu/libcuda.so.1", at 0x71218dd40545, in 
#8    Object "/lib/x86_64-linux-gnu/libcuda.so.1", at 0x71218e0e1885, in 
#7    Object "/lib/x86_64-linux-gnu/libcuda.so.1", at 0x71218e0e1663, in 
#6    Object "/opt/nvidia/nsight-systems/2022.1.3/target-linux-x64/libcupti.so.11.7", at 0x7121276fb83b, in 
#5    Object "/opt/nvidia/nsight-systems/2022.1.3/target-linux-x64/libcupti.so.11.7", at 0x7121276f1ace, in 
#4    Object "/opt/nvidia/nsight-systems/2022.1.3/target-linux-x64/libcupti.so.11.7", at 0x712127737fc8, in 
#3    Object "/opt/nvidia/nsight-systems/2022.1.3/target-linux-x64/libcupti.so.11.7", at 0x7121277093b4, in 
#2    Object "/opt/nvidia/nsight-systems/2022.1.3/target-linux-x64/libcupti.so.11.7", at 0x712127708fbf, in 
#1    Object "/opt/nvidia/nsight-systems/2022.1.3/target-linux-x64/libcupti.so.11.7", at 0x71212771d0e3, in 
#0    Object "/lib/x86_64-linux-gnu/libc.so.6", at 0x712196297ef4, in pthread_mutex_lock

System info:
Ubuntu 22.04
Driver Version: 550.90.07
CUDA Version: 12.4
NVIDIA Nsight Systems version 2022.1.3.3-1c7b5f7

2

First of all, please update to to 2024.5, your version is over 2.5 years old and there are innumerable improvements in the tool.

Secondly, I am going to suggest that you not use “nsys nvprof” that option was designed to be a temporary shortcut to allow users who were used to nvprof to get as close as possible to the options that that tool used to automatically run. I would instead just start with “nsys profile my_app”

3

Also, what is the GPU you are using?
Can you send the results of “nsys status -e”

4

@hwilper Ah you were right, it was a bit old, I just updated and it ran through my app.
I do miss the nvprof profiler and the simplicity of its report. But now I also have the option to explore a bit more the new features of nsys.
Thanks for the quick answer.

5

If you use the --stats option you will get a summary report that includes most of the same things.