Xavier NX GPU Code Latency


I have a GPU code of mine working on the MX450 GPU (laptop) at maximum 20 milliseconds. For 100 features, it gives 15 milliseconds.

When I compile and run the same code for Jetson Xavier NX, I get about 25 milliseconds latency at maximum but most of the cases the code runs about 5-6 milliseconds even if the feature count is large (about 20 features). Sometimes, with the same feature count, it can give about 25 milliseconds latency (for 20 features) .

On Jetson side, the code takes the real time camera images and runs the algorithm parallel on GPU.

Is the variation of the latency normal?
Any suggestions?


Best regards,

What kidnd of GPU code? Based on CUDA or ?

Did you boost the system by nvpmodel and jetson_clocks?
sudo nvpmodel -m 0
sudo jeston_clocks

Pease try kayccc’ ssuggestion and you can also execute sudo tegrastats to get loading of GPU.

The code is based on cuda, yes.

I tried with sudo nvpmodel -m2 and sudo jetson_clocks --fan.

Hi kayccc,

I tried your suggestion, it worked better. Thanks.

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