Install NVIDIA Drivers and Vulkan on Ubuntu

Installation Prerequisites

Upgrade the system and the kernel, then the machine if needed.

sudo apt update

sudo apt upgrade -y
sudo reboot

Install Kernel Headers

Install kernel headers and development packages.

sudo apt install linux-headers-$(uname -r)

Install the extra packages.

sudo apt install pciutils

Installing Vulkan Library

The rendering engine of HEAVY.AI (present in Enterprise Editions) requires a Vulkan-enabled driver and the Vulkan library. Without these components, the database itself may not be able to start.

Install the Vulkan library and its dependencies using apt.

sudo apt install libvulkan1

For more information about troubleshooting Vulkan, see the Vulkan Renderer section.

Installing NVIDIA Drivers

Installing NVIDIA drivers with support for the CUDA platform is required to run GPU-enabled versions of HEAVY.AI.

You can install NVIDIA drivers in multiple ways, we've outlined three available options below. If you would prefer not to decide, we recommend Option 1.

• Option 1: Install NVIDIA drivers with CUDA toolkit from NVIDIA Website

• Option 2: Install NVIDIA drivers via .run file using the NVIDIA Website

• Option 3: Install NVIDIA drivers using APT package manager

It is advisable to keep a record of the installation method used, as upgrading NVIDIA drivers will require the utilization of the same method for successful results.

What is CUDA? What is the CUDA toolkit?

CUDA is a parallel computing platform and application programming interface (API) model. It uses a CUDA-enabled graphics processing unit (GPU) for general-purpose processing. The CUDA platform provides direct access to the GPU virtual instruction set and parallel computation elements. For more information on CUDA unrelated to installing HEAVY.AI, see https://developer.nvidia.com/cuda-zone.

The CUDA Toolkit from NVIDIA provides everything you need to develop GPU-accelerated applications. The CUDA Toolkit includes GPU-accelerated libraries, a compiler, development tools and the CUDA runtime. The CUDA Toolkit is not required to run HEAVY.AI, but you must install the CUDA toolkit if you use advanced features like C++ User-Defined Functions and or User-Defined Table Functions to extend the database capabilities.

Option 1: Install NVIDIA Drivers with CUDA Toolkit from NVIDIA Website

Open https://developer.nvidia.com/cuda-toolkit-archive and select the desired CUDA Toolkit version to install.

The minimum CUDA version supported by HEAVY.AI is 11.4. We recommend using a release that has been available for at least two months.

In the "Target Platform" section, follow these steps:

1. For "Operating System" select Linux

2. For Architecture" select x86_64

3. For "Distribution" select Ubuntu

4. For "Version" select the version of your operating system (20.04)

5. For "Installer Type" choose deb (network) **

6. One by one, run the presented commands in the Installer Instructions section on your server.

** You may optionally use any of the "Installer Type" options available.

If you choose to use the .run file option, prior to running the installer you will need to manually install build-essentials using apt and change permissions of the downloaded .run file to allow execution.

Option 2: Install NVIDIA Drivers via .run file using the NVIDIA Website

Install the CUDA package for your platform and operating system according to the instructions on the NVIDIA website (https://www.nvidia.com/download/index.aspx).

If you don't know the exact GPU model in your system run this command

lspci -v | egrep "3D|VGA*.NVIDIA" | awk -F '\[|\]' ' { print $2 } '

You'll get an output in the format Product Type, Series and Model

Tesla T4

In this example, the Product type is Tesla the Series is T (as Turing), and the model is T4.

1. Select the Product Type as the one you got with the command.

2. Select the correct Product Series and Product Type for your installation.

3. In the Operating System dropdown list, select Linux 64-bit.

4. In the CUDA Toolkit dropdown list, click a supported version (11.4 or higher).

5. Click Search.

6. On the resulting page, verify the download information and click Download

7. On the subsequent page, if you agree to the terms, right click on "Agree and Download" and select "Copy Link Address". You may also manually download and transfer to your server, skipping the next step.

8. On your server, type wget and paste the URL you copied in the previous step. Press enter to download.

Please check that the driver's version you are downloading meets the HEAVI.AI minimum requirements

Install the tools needed for installation.

sudo apt install build-essential

Change the permissions of the downloaded .run file to allow execution, and run the installation.

chmod +x NVIDIA-Linux-x86_64-*.run
sudo ./NVIDIA-Linux--x86_64-*.run

Option 3: Install NVIDIA drivers using APT

Install a specific version of the driver for your GPU by installing the NVIDIA repository and using the apt package manager.

Be careful when choosing the driver version to install. Ensure that your GPU's model is supported and that meets the HEAVI.AI minimum requirements

Run the command to get a list of the available driver's version

apt list nvidia-driver-*
Listing... Done

nvidia-driver-450/bionic-updates,bionic-security 460.91.03-0ubuntu0.18.04.1 amd64
nvidia-driver-450-server/bionic-updates,bionic-security 450.172.01-0ubuntu0.18.04.1 amd64
nvidia-driver-455/bionic-updates,bionic-security 460.91.03-0ubuntu0.18.04.1 amd64
nvidia-driver-460/bionic-updates,bionic-security 470.103.01-0ubuntu0.18.04.1 amd64
nvidia-driver-465/bionic-updates,bionic-security 470.103.01-0ubuntu0.18.04.1 amd64
nvidia-driver-470/bionic-updates,bionic-security 470.103.01-0ubuntu0.18.04.1 amd64
nvidia-driver-470-server/bionic-updates,bionic-security 470.103.01-0ubuntu0.18.04.1 amd64
nvidia-driver-495/bionic-updates,bionic-security 510.60.02-0ubuntu0.18.04.1 amd64
nvidia-driver-510/bionic-updates,bionic-security 510.60.02-0ubuntu0.18.04.1 amd64
nvidia-driver-510-server/bionic-updates,bionic-security 510.47.03-0ubuntu0.18.04.1 amd64

Install the driver version needed with apt

sudo apt install nvidia-driver-<version>

NVIDIA Driver Post-Installation steps

Reboot your system to ensure the new version of the driver is loaded

sudo reboot

Verify Successful NVIDIA driver installation

Run nvidia-smi to verify that your drivers are installed correctly and recognize the GPUs in your environment. Depending on your environment, you should see something like this to confirm that your NVIDIA GPUs and drivers are present.

If you see an error like the following, the NVIDIA drivers are probably installed incorrectly:

NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. 
Make sure that the latest NVIDIA driver is installed and running.

Review the installation instructions, specifically checking for completion of install prerequisites, and correct any errors.

Install Vulkan library

The rendering engine of HEAVY.AI requires a Vulkan-enabled driver and the Vulkan library. Without these components, the database itself can't even start without disabling the back-end renderer.

Install the Vulkan library and its dependencies using apt.

sudo apt install libvulkan1

For more information about troubleshooting Vulkan, see the Vulkan Renderer section.

Advanced Installation

You must install the CUDA toolkit and Clang if you use advanced features like C++ User-Defined Functions and or User-Defined Table Functions to extend the database capabilities.

Install CUDA Toolkit ᴼᴾᵀᴵᴼᴺᴬᴸ

If you installed NVIDIA drivers using Option 1 above, the CUDA toolkit is already installed; you may proceed to the verification step below.

Install the NVIDIA public repository GPG key.

distribution=$(. /etc/os-release;echo $ID$VERSION_ID | sed -e 's/\.//g')
sudo apt-key adv --fetch-keys \
https://developer.download.nvidia.com/compute/cuda/repos/$distribution/x86_64/3bf863cc.pub

Add the repository.

echo "deb http://developer.download.nvidia.com/compute/cuda/repos/$distribution/x86_64 /" \
| sudo tee /etc/apt/sources.list.d/cuda.list
apt update

List the available Cuda toolkit versions.

apt list cuda-toolkit-* | grep -v config

Listing...
cuda-toolkit-10-0/unknown 10.0.130-1 amd64
cuda-toolkit-10-1/unknown 10.1.243-1 amd64
cuda-toolkit-10-2/unknown 10.2.89-1 amd64
cuda-toolkit-11-0/unknown 11.0.3-1 amd64
cuda-toolkit-11-1/unknown 11.1.1-1 amd64
cuda-toolkit-11-2/unknown 11.2.2-1 amd64
cuda-toolkit-11-3/unknown 11.3.1-1 amd64
cuda-toolkit-11-4/unknown 11.4.4-1 amd64
cuda-toolkit-11-5/unknown 11.5.2-1 amd64
cuda-toolkit-11-6/unknown 11.6.2-1 amd64
cuda-toolkit-11-7/unknown 11.7.0-1 amd64

Install the CUDA toolkit using apt.

sudo apt install cuda-toolkit-<version>

Verification

Check that everything is working and the toolkit has been installed.

/usr/local/cuda/bin/nvcc --version

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2020 NVIDIA Corporation
Built on Mon_Nov_30_19:08:53_PST_2020
Cuda compilation tools, release 11.2, V11.2.67
Build cuda_11.2.r11.2/compiler.29373293_0

Install Clang ᴼᴾᵀᴵᴼᴺᴬᴸ

You must install Clang if you use advanced features like C++ User-Defined Functions and or User-Defined Table Functions to extend the database capabilities. Install Clang and LLVM dependencies using apt.

sudo apt install clang

Verification

Check that the software is installed and in the execution path.

clang --version
clang version 10.0.0-4ubuntu1 
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin

For more information, see C++ User-Defined Functions.