ver. 9.7.0: "nextcloud-aio-memories" doesn't have an access to Nvidia GPU and can't do HW transcoding

[shperrung]

Hi!

After Nextcloud-AIO update from ver 9.6.0 to 9.7.0 Memories doesn't use HW transcoding with NVENC acceleration. nvidia-smi command inside "nextcloud-aio-memories" container doesn't show a presence of Nvidia GPU. However, if I create external go-vod container based on the same image "radialapps/go-vod:latest" and set its address:port in Nextcloud-AIO Memories app, NVENC will work but very slowly. This external go-vod returns information about GPU in response on nvidia-smi.

Steps to reproduce

1. Install Nextcloud-AIO version 9.6.0
2. Setup HW transcoding on Nvidia GPU in Memories with success.
3. Update Nextcloud-AIO to version 9.7.0 and loose HW transcode in Memories.

Expected behavior

Nvidia GPU is accessible inside "nextcloud-aio-memories" container and adequate response on nvidia-smi. Transcoding videos with Nvenc capabilities in Memories app

Actual behavior

Memories app doesn't paly video when "acceleration with NVENC" is enabled. nvidia-smi command inside "nextcloud-aio-memories" container returns bash: nvidia-smi: command not found

Other information

Host OS

Debian 12 (Openmediavault 7)

Output of sudo docker info

`~$ docker info
Client: Docker Engine - Community
Version: 27.3.1
Context: default
Debug Mode: false
Plugins:
buildx: Docker Buildx (Docker Inc.)
Version: v0.17.1
Path: /usr/libexec/docker/cli-plugins/docker-buildx
compose: Docker Compose (Docker Inc.)
Version: v2.29.7
Path: /usr/libexec/docker/cli-plugins/docker-compose

Server:
Containers: 45
Running: 43
Paused: 0
Stopped: 2
Images: 51
Server Version: 27.3.1
Storage Driver: overlay2
Backing Filesystem: extfs
Supports d_type: true
Using metacopy: false
Native Overlay Diff: true
userxattr: false
Logging Driver: json-file
Cgroup Driver: systemd
Cgroup Version: 2
Plugins:
Volume: local
Network: bridge host ipvlan macvlan null overlay
Log: awslogs fluentd gcplogs gelf journald json-file local splunk syslog
Swarm: inactive
Runtimes: io.containerd.runc.v2 nvidia runc
Default Runtime: runc
Init Binary: docker-init
containerd version: 7f7fdf5fed64eb6a7caf99b3e12efcf9d60e311c
runc version: v1.1.14-0-g2c9f560
init version: de40ad0
Security Options:
apparmor
seccomp
Profile: builtin
cgroupns
Kernel Version: 6.8.12-2-pve
Operating System: Debian GNU/Linux 12 (bookworm)
OSType: linux
Architecture: x86_64
CPUs: 16
Total Memory: 62.72GiB
Name: omv6
ID: 6I4Q:3WBL:UVNA:MNFF:ZUYY:E6JI:UXLY:L7YC:2EY5:ODVU:UPQB:5HIZ
Docker Root Dir: /var/lib/docker
Debug Mode: false
Experimental: false
Insecure Registries:
127.0.0.0/8
Live Restore Enabled: false`

Docker run command or docker-compose file that you used

see attached aio_compose.txt

Other valuable info

[szaimen]

Hi, the community container for memories does indeed not support nvidia gpus yet afaik. This is missing first: #5132

[shperrung]

Thank you for clear answer.
For those, who still wants to utilize their nvidia, use external container with go-vod attached to "nextcloud-aio" docker network:

services:
  memories:
    image: radialapps/go-vod
    restart: always
    networks:
      - go-vod-ext
    environment:
      - NEXTCLOUD_HOST=https://aio.YOUR_DOMAIN.duckdns.org
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=all
      - GPU=all
    #devices:
      #- /dev/dri:/dev/dri
    volumes:
      - /docker/aiodata:/mnt/ncdata:ro
      - /mnt/pool:/mnt/pool:ro
    runtime: nvidia
networks:
  go-vod-ext:
    name: nextcloud-aio
    external: true

You will get second "go-vod" container named "go-vod-memories-1" based on the same image as "nextcloud-aio-memories" but it's powered with NVENC.
Use in Memories "external transcoder" settings this: "Connection address (same as bind if local): go-vod-memories-1:47788".

[priyankub]

One could also opt not to run the memories container through AIO, and instead run their separate nextcloud-aio-memories container with the go-vod image and all above options.
That's what I am doing for a seamless experience with memories able to use NVENC.
@szaimen - Could that be enabled natively as an option please, rather than users having to run their separate memories container

[szaimen]

It cannot as it breaks on instanced without gpu. See #4277