How to test NGC-Ready Server

Hardware:
B7102, BOS v1.03.B10, BMC v4.0
CPU: Xeon 6128x2
RAM: DDR4 128GB
OS: Ubuntu 18.04.2 LTS Server
GPU: Nvidia Tesla T4 x4

AP: Nvidia


Test Process

Install the Ubuntu Operating System and Nvidia Tesla Driver

1. Install Ubuntu 18.04.2 LTS server

2. Login as user



3. Download Nvidia 418.40 driver

4. Install Nvidia 418.40 driver local repository

5. add the local repo key

6. Prioritize the local repo over the network repos

7. cat local-cuda to verify

8. Update the apt metadata

9. Install Nvidia 418.40 driver

10. Confirm that you can see your Nvidia Tesla T4 card in the nvidia output

Installing Dcoker and the Docker Utility Engine for Nvidia GPUs

Enabling the Docker Repository

1. Installs the Docker prerequisites

2. Adds the Docker official GPG key

3. Add the official stable Docker repository

Installing the Nvidia Container Runtime for Nvidia GPUs

4. Reload docker
5. Test the Docker install and Nvidia container runtime by pulling the latest official CUDA image
and running nvidia-smi

Setting Up docker Options

1. Edits the contents of /etc/systemd/system/docker.service.d/override.conf.
2. Reload the systemd manager configuration.
3. Reload Docker.



Test NGC Sanity

1. extract NGC-Ready-master to Linux
2. type cd /tf to change folder to tf
3. vi lanuch to add red parameters

docker build --network=host --no-cache -t $TESTNAME

docker run --network=host -it --security-opt

docker run --network=host -it --runtime=nvidia

4. type ./launch to execute script
note: this script may take over 8 hours.