Add ability to run NVIDIA containers without the NVIDIA container toolkit

docs/compose-files/docker-compose.default.yml

@@ -36,6 +36,14 @@ services:
       - /dev/uinput
       # Add AMD/Intel HW accelerated video encoding/decoding devices (optional)
       #- /dev/dri
+      # Add NVIDIA HW accelerated devices (optional - must be run with runc docker runtime. If you use the nvidia runtime, this is not needed)
+      #- /dev/nvidia0
+      #- /dev/nvidiactl
+      #- /dev/nvidia-modeset
+      #- /dev/nvidia-uvm
+      #- /dev/nvidia-uvm-tools
+      #- /dev/nvidia-caps/nvidia-cap1
+      #- /dev/nvidia-caps/nvidia-cap2
     # Ensure container access to devices 13:*
     device_cgroup_rules:
       - 'c 13:* rmw'

docs/ubuntu-server.md

@@ -24,6 +24,9 @@ Follow the [official instructions](https://docs.nvidia.com/datacenter/cloud-nati
 
 If you do this, ensure that when you configure Docker Compose in the next step you choose the `nvidia` runtime.
 
+Alternately, it is possible to run the container without the NVIDIA runtime by uncommenting the `/dev/nvidia` devices in the Compose file.
+
+
 ## CONFIGURE DOCKER COMPOSE:
 
 Once you have installed docker, follow the [Compose Files](./docker-compose.md) section and select the right configuration file for your hardware.