We have downloaded and installed the newly released DS 4.0 on a server with two T4 cards. The installation is smooth following the Guidance. However when trying to run the deepstream-app and deepstream-test1-app, these error occurred after pipeline is set to running:
ERROR from sink_sub_bin_encoder1: Device '/dev/nvhost-msenc' failed during initialization
Debug info: gstv4l2object.c(4050): got_v4l2_object_set_format_full (): /GstPipeline:pipeline/GstBin:processing_bin_0/GstBin:sink_bin/GstBin:sink_sub_bin1/nvv4l2h265enc:sink_sub_bin_encoder1:
Call to S_FMT failed for YM12 @ 1280x720: Unknown error -1
and this repeats a few times before the application quits.
We tried both H264 and H265 encoding, and switched between video file src and live RTSP stream, getting the same error.
So we tried again with gst-launch-1.0, copied the command from the FAQ, just got the same result. However, when replacing the sink with gst’s fakesink, the pipeline works (though nothing is displayed, of course). Disabled all sinks in config file of deepstream-app also results like this, running pipeline without error nor video stream, by using a default fakesink.
The dumped .dot graph shows the encoder nvv4l2h264enc tried to access device /dev/nvhost-msenc without success, just like the error above complained. It looks like the hardware encoder on T4.
But ls /dev/ cannot find such device, nor any other /dev/nvhost-* devices. A quick search reports such devices are used in Jetson, but no result in Tesla webpages.
After re-installing the latest Nvidia driver from official website (and reboot), the problem seems gone.
Both driver versions are 418.67, we don’t know the exact difference since previous installation is gone.
There is still no /dev/nvhost-msenc listed, but the program no longer complains about it.
Also the device="/dev/nvhost-msenc" property can be seen in dumped .dot graph, in sink_sub_bin_encoder1 of type nvv4l2h265enc as before.
Well. After installing, rebooting, re-installing on host ubuntu everything I was able to run samples without issues. Not in docker container. I’ll test docker image a bit later.
Receiving the same set of issues with deep stream container 4.0-19.07, whenever I enable sink[1] to write rendered out.mp4 file it gives me this error on Tesla T4.