ros_rtsp
ros_rtsp copied to clipboard
nvidia codec
Hi,
I am working to use GPU-accelerated encoder instead as xh264enc causes considerably high CPU usage. What I am currently doing is simply replacing xh264enc
with nvh264enc
in the pipeline string and removing those properties not supported by nvh264enc like tune=zerolatency
etc. I can see a substantial drop of CPU usage on the sender side, but the receiver side gets higher latency. How could I reconstruct the gstreamer pipeline properly to achieve the same low latency as the original xh264enc? There seems much fewer options that can be used to tune nvh264enc.
I came across the same problem. I was able to get considerably less latency using CPU and x265 than a GPU. Might be a driver issue, or maybe something is double handling the memory.
Sorry but I no longer have time to work on this further.
You can have a look at what corenel did in his fork? Let me know how it goes.
https://github.com/corenel/ros_rtsp