Hi All,
We have tested our camera driver and application in L4T28.1.
But streaming performance is very slow compared to L4T24.2.1(almost 10 times slower).
In comparison with L4T24.1, L4T28.1 takes:-
- V4L2 capture time (increased ~4 times)
- CUDA ISP processing time (increased 9 times)
- Rendering time (increased ~7 times)
We executed jetson_clocks.sh and maxPerf.sh in 28.1. But errors occured.
Any suggestions?
Thanks inadvance,
Mathews
Hi mats_X1,
Sorry that I don’t get your statistics. Do you mean all latency increasing higher than “4 times” in each steps?
Could you share the real values here? Also, it would be better if you could share your codes.
Hi WayneWWW,
Thanks for your mail.
Compared to 24.2.1, in 28.1 latency has increased
In 24.2.1,
Capture time : 0.06 ms
CUDA ISP processing time : 40ms
Render time : 13ms
In 28.1,
Capture time : 0.21ms
CUDA ISP processing time : 360ms
Render time : 90ms
Sorry,I can’t share source code.
Regards,
Mathews
Hi, can you break down CUDA ISP processing and see which API takes longer time?
hello mats_X1,
may i know your sensor frame rate?
also, could you reproduce the “Capture time” latency of R24.2.1 and R28.1 by v4l2-ctl?
thanks