Closed plasticfist closed 1 month ago
Thanks for the detailed report. Will see if upgrading vaapi solves the issue, will let you know when its ready to test
Edit: Actually this might be harder than i thought. The drivers are installed from apt
and the container runs Ubuntu 20.04
Upgrading the base image is not an option right now due to other dependencies not working in later Ubuntu versions.
Do you have a good idea of how this could be done?
The intel driver, GmmLib, LibVA can be compiled from source in principle.
What exactly are the dependencies that prevent you from updating to ubuntu 22.04, for example?
Ubuntu 20.04 is already a fairly old distribution and similar problems will appear more and more often... With this version of VAAPI and the iHD_drv_video driver, I am experiencing memory leaks on the Celeron J4105, and I had to switch to i965_drv_video.
Building from source is definitely preferable! However doing that is pretty time consuming in my experience. Do you know how to do it or know of any good guides?
The dependency issues are mainly Python related, for instance Pycoral does not officially support Python 3.10
You can use 3rd party repos to install python, ppa:deadsnakes/ppa at example have some versions:
Great idea, will try that!
I am experiencing the same problem. Both on N100 and i5-12600k.
Will try to fast track some upgrades of dependencies before v3.0.0
I have successfully updated to Ubuntu 22.04 and have a PoC for running the EdgeTPU component under python 3.9 in a spawn subprocess.
Just need to clean up the code a bit before pushing a new beta
FWIW, the VA-API Issue still exists on the latest dev build for me. Also using Intel N100 with SR-IOV.
I edited the file /etc/apt/sources.list.d/intel.gpu.focal.list
to be jammy instead of focal and installed intel-media-va-driver-non-free
again. Now vainfo is reporting my GPU and after restarting the viseron container, VA-API is recognized!
****** Checking for hardware acceleration platforms ******
OpenCL is available!
VA-API is available!
CUDA cannot be used
Thanks for troubleshooting! Will fix soon. Hopefully I can release the next beta in a few days
Hit a small snag, i upgraded OpenVINO in the process to version 2023.3 since older versions are not available for Ubuntu 22.04. This version is causing segfaults when running object detection.
Have to revert back to an old version and do some testing before i can release the beta
I decided to ignore the OpenVINO error for now and added a note to the docs that it is broken. Hopefully OpenCV 4.10.0 fixes this issue when it is released.
I just pushed 3.0.0b4
, builds should be done in a few minutes.
Please give it a go when you get the chance!
VA-API working fine for me with 3.0.0b4 and Intel N100 using SR-IOV!
Great to hear! Thanks for the quick response
Closing as this is addressed in the latest beta. Please create a new issue if you encounter any problems
inside the viseron container va-api is broken. I noticed a thread about an issue here was solved by adding more drivers...
Note that maybe a clue: I notice that viseron is reporting VA-API version 1.15.0, and my working container reports VA-API version 1.17.0
The server is an N100 CPU (low power), graphics are SRIOV virtualized through proxmox into a VM running docker containers, a PCI Google Coral TPU passed in, and I think viseron is using the TPU?
host OS
SRIOV references
viseron logs at start up are
inside the viseron container, vainfo says
compared this to what vainfo reports inside the frigate container, same server and settings