Hello everyone,
we are trying to run a five camera setup on the Jetson Xavier.
The specs are as follows:
3x ~11.8 MP
2x 8 MP
45fps
As we preferably want to use the ISP of the Xavier, we tested the setup with the nvarguscamerasrc.
The results showed an average framerate of less than 45 fps. It was moving around 32 fps.
We verified, that if we use a pipeline with v4l2-ctl, we can achieve the desired framerate for all cameras. In single operational mode as well as all cameras in parallel.
For all tests we were using the MAXN nvpmodel and activated jetson_clocks. We also boosted the clk values for vi, isp and nvcsi to the value in max_rate, as some answers in other posts proposed.
Our conclusion was, that the ISP cannot handle such a high throughput. Our research yielded contradictory results, as some sources claimed a throughput of 1.5GPix/s, whereas some official NVIDIA presentations claimed 2.4 GPix/s.
Can someone confirm one of those values, or is the real number not officially known?
Maybe the problem is also elsewhere, but this was our initial guess. So any hints are welcome.
Best regards and thanks for any help,
Markus