The intel k8s device plugin doc includes a workaround for working in a multi-gpu environment. The script (render-device.sh) will return the correct intel render device to use if you're oversubscribing a gpu or if you have more than one gpu in a device. Is there an opportunity to use render-device to automatically select the correct render device and replace it in the config file?
Without support Frigate may fail to run if a node has more than one GPU (discrete gpu and intel) or if you've allowed oversubscription in the device plugin.
The built-in environment variable replacement in Frigate doesn't allow replacement in the hwaccel_args section, so the chart could use an init container with shared volume for config to make the replacement.
The intel k8s device plugin doc includes a workaround for working in a multi-gpu environment. The script (render-device.sh) will return the correct intel render device to use if you're oversubscribing a gpu or if you have more than one gpu in a device. Is there an opportunity to use render-device to automatically select the correct render device and replace it in the config file?
Without support Frigate may fail to run if a node has more than one GPU (discrete gpu and intel) or if you've allowed oversubscription in the device plugin.
The built-in environment variable replacement in Frigate doesn't allow replacement in the hwaccel_args section, so the chart could use an init container with shared volume for config to make the replacement.
https://github.com/intel/intel-device-plugins-for-kubernetes/blob/main/cmd/gpu_plugin/render-device.sh