Memory overrun

Developer Tools Nsight DL Designer
1

Trying to load a detector network in DL Designer causes insanely high ram usage. There needs to be some cache limiting because I ran out of memory on a system with 64 gigs of RAM trying to load a single object detector ONNX file.

2

Hi, sorry for your trouble. To help us better understand and investigate the problem, could you share more details?

  • What is your system, Windows, Linux, or L4T?
  • In what circumstance did the memory overrun happen? loading a model, editing a model, profiling a model (with which profiler)?
  • Is it possible for you to share with us your model to reproduce the problem? If not, can you tell us more about the model itself?

Thanks!
Gaoyan Xie

3

Hey,

Thanks for the quick reply.

I’m on Windows 11.

The memory overrun happens if I try to load an onnx model into the DL Designer app by dragging the file into the window. The ambient memory before triggering hovers at around ~8 GB and then shoots up to 64+ and crashes when trying to load the model.

It is a MMRotate based detector model. However, I probably cannot share the model as is, as it has proprietary elements. However, I will have a talk with the team to see how much I can divulge and get back to you. Would any other information be helpful?

4

Thanks, our engineering team is actively investigating this problem.

Gaoyan Xie

5

Hi,

This problem has been fixed in our latest release (2025.1): Nsight Deep Learning Designer - Get Started | NVIDIA Developer

Please give it a try and let us know if the problem is resolved on your end.

Thanks,
Gaoyan Xie

6

Hey, yes it is fixed now. The overrun is no longer present for the same action.