Nsys memory usage

I have profiled 3d-unet with nsys and the qdstrm file is 7.3GB. As you can see in the picture below, when I import it with 2022.2.1, it eats all 128 GB of memory plus 32 GB of swap. I think there is a memory leak there. I don’t know how to share a sample for that. But I can share the whole 7.3GB file.

Hi @mahmood.nt , may I ask, for how long does the profiled application run?

Running 3d-unet with nsys takes about 4.5 hours on my 3080 device.

Thanks for your response. In this case I think the amount of collected data is quite large.

As stated in the tool’s documentation:

Nsight Systems will collect a different amount of data/sec depending on options, but in general Nsight Systems does not support runs of more than 5 minutes duration.

Can you try using the CLI switches --duration and/or --delay to narrow down the collection in the area of the application that you want to focus you analysis? There are equivalent options if you are profiling from the GUI, “Start profiling after”, “Limit profiling to”.

Another alternative to reduce the amount of collected data is to enable less tracing options or disable CPU sampling. Depending on what data you are most interested collecting.