GPU transcoding with Emby / Plex using docker-nvidia | Funky Penguin


#1

Recently I helped a Patron to setup an instance of my Emby docker-swarm recipe, but with the extra bonus of having all transcoding done using his GPU. Note that this would work equally well for Plex, or any other “Dockerized” application which would, technically, support GPU processing.


This is a companion discussion topic for the original entry at https://www.funkypenguin.co.nz/note/gpu-transcoding-with-emby-plex-using-docker-nvidia/

#2

I have two Windows 10 desktops in my home. One has a NVIDIA GTX1060 and the other has an AMD R470.

Does AMD have a similar setup so both hosts can transcode properly?

Also, with this setup, would you recommend I run docker on a VM running linux? Or just run Docker for windows?

Thanks.


#3

Sorry, missed this. Best bet here is running the GTX under nvidia-docker on Linux. I don’t think it’s at all possible in Windows :wink:


#4

Hey @funkypenguin,

I’m trying out this guide and I think I’ve got everything set correctly. I have the runtime as nvidia and the env variable set for plex. I can see the Nvidia card in the plex container but I’m getting the error 11230:Sep 16, 2018 20:15:00.432 [0x7f1b733ff700] DEBUG - Codecs: hardware transcoding: opening hw device failed - probably not supported by this system, error: Invalid argument when the server is running.

Any ideas?


#5

Hey @lyoko37,

What’s the result of df and env if you’re shelled into the container?

D


#6

Thanks for the quick reply @funkypenguin

Filesystem       1K-blocks        Used   Available Use% Mounted on
overlay           49272632    30685544    16061136  66% /
tmpfs                65536           0       65536   0% /dev
tmpfs             32931428           0    32931428   0% /sys/fs/cgroup
storage        42890405376 19690219520 23200185856  46% /movies
/dev/sdi1         49272632    30685544    16061136  66% /transcode
shm                  65536           8       65528   1% /dev/shm
tmpfs             32931428          12    32931416   1% /proc/driver/nvidia
udev              32902140           0    32902140   0% /dev/nvidia0
tmpfs             32931428           0    32931428   0% /proc/acpi
tmpfs             32931428           0    32931428   0% /proc/scsi
tmpfs             32931428           0    32931428   0% /sys/firmware

root@TheHoard:/dev# env
HOSTNAME=TheHoard
PUID=1000
TERM=xterm
LC_ALL=C.UTF-8
PGID=1000
NVIDIA_VISIBLE_DEVICES=all
PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
PWD=/dev
LANG=C.UTF-8
SHLVL=1
HOME=/config
affinity:container==351dadd5522594140ce25ef75aa43c46cd10c375ea2f91f29211ebe251da9b87
CHANGE_CONFIG_DIR_OWNERSHIP=true
_=/usr/bin/env
OLDPWD=/

Inside the container I also see:

root@TheHoard:/dev# ls
core  fd  full	mqueue	null  nvidia0  nvidiactl  ptmx	pts  random  shm  stderr  stdin  stdout  tty  urandom  zero

root@TheHoard:/dev# nvidia-smi
Sun Sep 16 21:32:45 2018
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 390.87                 Driver Version: 390.87                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  Quadro P2000        Off  | 00000000:02:00.0 Off |                  N/A |
|  0%   27C    P0    16W /  75W |      0MiB /  5057MiB |      3%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+
root@TheHoard:/dev#

It seems like the card is showing up in the container there but I’m not sure.


#7

Just for reference as well my docker compose file looks like this:

  plex:
    image: plexinc/pms-docker:plexpass
    container_name: plex
    runtime: nvidia
    environment:
      - PGID=1000
      - PUID=1000
      - NVIDIA_VISIBLE_DEVICES=all
    volumes:
      - /storage/container_configs/plex:/config
      - /storage/media/tv_shows:/tvshows
      - /storage/media/movies_4k:/tvshows_4k
      - /storage/media/movies:/movies
    network_mode: "host"
    restart: always

#8

Aha.
Try adding this additional env variable:

NVIDIA_DRIVER_CAPABILITIES=compute,utility

#9

Hmm

Still seeing this when it attempts to transcode:

Sep 16, 2018 21:53:16.412 [0x7f8d9dfff700] DEBUG - Codecs: hardware transcoding: testing API vaapi
Sep 16, 2018 21:53:16.412 [0x7f8d9dfff700] DEBUG - Codecs: hardware transcoding: opening hw device failed - probably not supported by this system, error: Invalid argument
Sep 16, 2018 21:53:16.414 [0x7f8d9dfff700] DEBUG - Codecs: hardware transcoding: testing API vaapi
Sep 16, 2018 21:53:16.414 [0x7f8d9dfff700] DEBUG - Codecs: hardware transcoding: opening hw device failed - probably not supported by this system, error: Invalid argument

Updated compose file:

  plex:
    image: plexinc/pms-docker:plexpass
    container_name: plex
    runtime: nvidia
    environment:
      - PGID=1000
      - PUID=1000
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
    volumes:
      - /storage/container_configs/plex:/config
      - /storage/media/tv_shows:/tvshows
      - /storage/media/movies_4k:/tvshows_4k
      - /storage/media/movies:/movies
    network_mode: "host"
    restart: always

#10

Here’s the output from a working Emby container with an nVidia GPU exposed:

/ # df
Filesystem           1K-blocks      Used Available Use% Mounted on
overlay              3914269104 2476888104 1238524504  67% /
tmpfs                    65536         0     65536   0% /dev
tmpfs                 32964632         0  32964632   0% /sys/fs/cgroup
/dev/md1             3914269104 2476888104 1238524504  67% /config
prod1:/storage       242960993280 216413934592 21663166464  91% /media
/dev/md1             3914269104 2476947852 1238464756  67% /etc/resolv.conf
/dev/md1             3914269104 2476947884 1238464724  67% /etc/hostname
/dev/md1             3914269104 2476947924 1238464684  67% /etc/hosts
shm                      65536         0     65536   0% /dev/shm
tmpfs                 32964632        12  32964620   0% /proc/driver/nvidia
/dev/md1             3914269104 2476948060 1238464548  67% /usr/bin/nvidia-smi
/dev/md1             3914269104 2476948108 1238464500  67% /usr/bin/nvidia-debugdump
/dev/md1             3914269104 2476948148 1238464460  67% /usr/bin/nvidia-persistenced
/dev/md1             3914269104 2476948188 1238464420  67% /usr/bin/nvidia-cuda-mps-control
/dev/md1             3914269104 2476948236 1238464372  67% /usr/bin/nvidia-cuda-mps-server
/dev/md1             3914269104 2476948268 1238464340  67% /usr/lib64/libnvidia-ml.so.396.54
/dev/md1             3914269104 2476948316 1238464292  67% /usr/lib64/libnvidia-cfg.so.396.54
/dev/md1             3914269104 2476948352 1238464256  67% /usr/lib64/libcuda.so.396.54
/dev/md1             3914269104 2476948396 1238464212  67% /usr/lib64/libnvidia-opencl.so.396.54
/dev/md1             3914269104 2476948444 1238464164  67% /usr/lib64/libnvidia-ptxjitcompiler.so.396.54
/dev/md1             3914269104 2476948492 1238464116  67% /usr/lib64/libnvidia-fatbinaryloader.so.396.54
/dev/md1             3914269104 2476948540 1238464068  67% /usr/lib64/libnvidia-compiler.so.396.54
/dev/md1             3914269104 2476948584 1238464024  67% /usr/lib64/libvdpau_nvidia.so.396.54
/dev/md1             3914269104 2476948620 1238463988  67% /usr/lib64/libnvidia-encode.so.396.54
/dev/md1             3914269104 2476948668 1238463940  67% /usr/lib64/libnvcuvid.so.396.54

And environment:

/ # env | grep NVIDIA
NVIDIA_DRIVER_CAPABILITIES=compute,utility,video
NVIDIA_VISIBLE_DEVICES=all
/ #

#12

I don’t think so. nvidia-docker adds an additional runtime to your docker environment, and it’s unLIKELY this’ll work with UnRAID :slight_smile:


#13

I only see (hw) appear on the encoding side and not decoding for Plex on ubuntu 16.04. Is this a known issue or did I do something wrong?

EDIT: Seems to be a limitation of the plex transcoder that should eventually get updated. https://forums.plex.tv/t/hardware-accelerated-decode-nvidia-for-linux/233510


#14

Thank you for this mini tutorial, it really helped me out getting my own Emby container running with nvenc!


#15

You’re welcome :slight_smile: FYI, there’s a bug in the current stable version which occasionally disables your transcoding settings. It’s fixed in the recent betas, and will be addressed in stable 3.6.0.