I use 32G AGX Orin, jetpack 5.0.2.
I need use AGX Orin to encode 4 1080p camera to h265 in parallel, and average encode delay must small than 10ms for every frame.
So I add time delay measurement code in 15_multivideo_encode:
And use command below to speed up nvcode and vic frequency:
echo "793600000" > /sys/class/devfreq/154c0000.nvenc/userspace/set_freq echo "userspace" > /sys/class/devfreq/15340000.vic/governor echo "729600000" > /sys/class/devfreq/15340000.vic/userspace/set_freq
Then use command below to run multivideo_encode
./multivideo_encode num_files 4 ./video_test.yuv 1920 1080 H265 sample_outdoor_car_1080p_10fps.h265 ./video_test.yuv 1920 1080 H265 sample_outdoor_car_1080p_10fps_second.h265 ./video_test.yuv 1920 1080 H265 sample_outdoor_car_1080p_10fps_third.h265 ./video_test.yuv 1920 1080 H265 sample_outdoor_car_1080p_10fps_forth.h265 -br 2000000 -hpt 1 -p main -l high4.0
Then print log below:
From the log, it shows all thread delay is much more than 10ms. And seems I frame encode cost maximum delay.
So if it possible to speed up encode process? What is the banchmark of nvenc in 32G AGX Orin? Will 64G AGX Orin will better？