Hello everyone. I`ve obsered wierd thing when using gstreamer. under ubuntu 24.04LTS. Im trying to decode and display UDP raw H265 live stream (1080p, 60fps) from ultra low latency camera with lowest latency as it is possible. Here is my pipeline:
gst-launch-1.0 udpsrc port=5600 ! tee name=videoTee ! queue ! tee name=t t. ! queue ! application/x-rtp,payload=97, clock-rate=90000, encoding-name=H265 ! rtpjitterbuffer latency=0 ! rtph265depay ! mpegtsmux name=ts ! filesink location=/home/record-$(date +%y%m%d_%H%M%S).tsn sync=false t. ! queue leaky=1 ! tee name=audioTee ! queue ! application/x-rtp, payload=98, encoding-name=OPUS ! rtpjitterbuffer latency=0 do-lost=true drop-on-latency=true ! rtpopusdepay ! ts. audioTee. ! queue leaky=0 ! application/x-rtp, payload=98, encoding-name=OPUS ! rtpjitterbuffer latency=0 ! rtpopusdepay ! opusdec ! audioconvert ! audioresample ! pipewiresink blocksize=128 mode=render processing-deadline=0 sync=false async=false videoTee. ! queue ! application/x-rtp,payload=97, clock-rate=90000, encoding-name=H265 ! rtpjitterbuffer latency=0 ! rtph265depay ! vaapih265dec ! fpsdisplaysink fps-update-interval=200 video-sink=xvimagesink sync=false
Im comparing two hardware scenarios:
1 - MSI laptop, model THIN B12UCX intel core i5 12450H with RTX2050
2 - DELL 3050 micropc intel core i5 6500T with intel HD 530
On MSI laptop when using vaapih265dec decoding goue to intel GPU and the image is very fast and very realistic, when I switch to nvidia decoder nvh265dec it feels little less realistic but latency seems simillar. The same situation looks when I swap to Dell 3050 with intel 530 gpu. Latency looks simillar but view is less realistic, maybe less nervous which seems to be a little bigger latency but when merging with timer difference is maybe 5-10ms max. I`m uing this same pipeline for both intel gpus. It looks like the new intel GPU from i5 12450H works faster than anything else what I was using before for decoding or maybe I can get the same effect from any other hardware acelerated pipeline but with some changes of pipeline ? If it is because of GPU perfomance why it is not the same with RTX gpu ?