Getting started with Multi-Instance GPU (MIG) on the NC A100 v4 series

Published Wednesday, August 3, 2022

By Hugo Affaticati, Program Manager

Useful resources

Information on the NC A100 v4-series: Microsoft

Information on Multi-Instance GPU (MIG): NVIDIA

Pre-requisites

Deploy a virtual machine on Microsoft Azure Portal.

Key values

Size: NC24ads A100 v4 (also available NC48ads A100 v4 and NC96ads A100 v4)
Image: Ubuntu HPC 18.04 (recommended, also available Ubuntu HPC 20.04)
Availability: no redundancy required for benchmarking

Step 1: NVIDIA driver and CUDA

Verify the NVIDIA driver version:

cd /mnt
nvidia-smi

If the driver version is less than 510, update both the driver and the CUDA versions

sudo wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-ubuntu1804.pin 

sudo mv cuda-ubuntu1804.pin /etc/apt/preferences.d/cuda-repository-pin-600

sudo wget https://developer.download.nvidia.com/compute/cuda/11.6.1/local_installers/cuda-repo-ubuntu1804-11-6-local_11.6.1-510.47.03-1_amd64.deb 

sudo dpkg -i cuda-repo-ubuntu1804-11-6-local_11.6.1-510.47.03-1_amd64.deb

sudo apt-key add /var/cuda-repo-ubuntu1804-11-6-local/7fa2af80.pub

sudo apt-get update

sudo apt-get -y install cuda

Restart the machine

sudo reboot

Step 2: Docker

The next step is to update Docker to the latest version.

cd /mnt

sudo apt update

sudo apt-get install apt-transport-https ca-certificates curl software-properties-common

curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo apt-key add -

sudo add-apt-repository "deb [arch=amd64] https://download.docker.com/linux/ubuntu  $(lsb_release -cs)  stable"

sudo apt update

sudo apt-get install docker-ce

Update the Docker root directory in the docker daemon configuration file

sudo vi /etc/docker/daemon.json

Add the line after the first curly bracket

"data-root": "/mnt/resource_nvme/data",

Verify the previous steps and enable docker

docker --version

sudo systemctl restart docker

sudo systemctl enable docker

sudo groupadd docker

sudo usermod -aG docker $USER

newgrp docker

You should not have any permission issues when running

docker info

Step 3: Enable MIG

Enable MIG mode

sudo nvidia-smi -mig 1

You may have to reboot the machine after this command. Then, verify that the MIG mode is enabled:

nvidia-smi

Create seven GPU instance IDs and the compute instance IDs:

sudo nvidia-smi mig -cgi 19,19,19,19,19,19,19
sudo nvidia-smi mig -cci

For two or three MIG instances you can use respectively:

sudo nvidia-smi mig -cgi 9,9
sudo nvidia-smi mig -cci

sudo nvidia-smi mig -cgi 14,14,14
sudo nvidia-smi mig -cci

Display the GPU instance profiles:

sudo nvidia-smi mig -lgip

Get the list of the MIG devices

nvidia-smi -L

Step 4: Mount the NVMe disks

cd /mnt
sudo vi nvme.sh

Copy and paste the following mounting script:

#!/bin/bash 

NVME_DISKS_NAME=`ls /dev/nvme*n1`
NVME_DISKS=`ls -latr /dev/nvme*n1 | wc -l`

echo "Number of NVMe Disks: $NVME_DISKS"

if [ "$NVME_DISKS" == "0" ]
then
    exit 0
else
    mkdir -p /mnt/resource_nvme
    # Needed incase something did not unmount as expected. This will delete any data that may be left behind
    mdadm  --stop /dev/md*
    mdadm --create /dev/md128 -f --run --level 0 --raid-devices $NVME_DISKS $NVME_DISKS_NAME
    mkfs.xfs -f /dev/md128
    mount /dev/md128 /mnt/resource_nvme
fi

chmod 1777 /mnt/resource_nvme