Install the GRID driver (Linux) - Elastic GPU Service - Alibaba Cloud Documentation Center

In AI computing and graphics-accelerated scenarios such as OpenGL, Direct3D, and cloud gaming, the GRID driver enables high-performance GPU computing and smoother graphics rendering. You can use Cloud Assistant to quickly install the GRID driver.

Note

Review the Overview of Cloud Assistant before proceeding. The Cloud Assistant agent is pre-installed on vGPU-accelerated instances.

Supported instances

Cloud Assistant supports GRID driver installation only on Linux instances of the sgn8ia, vgn6i-vws, vgn7i-vws, and sgn7i-vws vGPU-accelerated instance families (vgn/sgn series). The following table lists supported GRID driver and CUDA versions.

Instance family

Public image version

GRID driver version

CUDA version

sgn8ia

Alibaba Cloud Linux 2/3
Ubuntu 22.04/20.04/18.04
CentOS 8.x/7.x

535.183.06

12.2

vgn6i-vws,

vgn7i-vws, and sgn7i-vws

Alibaba Cloud Linux 2/3
Ubuntu 22.04/20.04/18.04
CentOS 8.x/7.x

470.256.03

11.4

Important

Linux GPU-accelerated compute-optimized instances such as gn7i, gn6i, ebmgn7i, and ebmgn6i do not require a GRID driver. The standard NVIDIA driver is sufficient. GPU-accelerated compute-optimized instances.

Procedure

Log on to the ECS console.
In the left-side navigation pane, choose Maintenance & Monitoring > Cloud Assistant.
In the upper-left corner, select the region of the target instance.
On the ECS Instance tab, find the target instance and click Run Command in the Actions column.
In the Create Command pane, configure the parameters in the Command Information section.

Set the following parameters and keep the defaults for the rest. Full parameter descriptions: Run a command on an instance.

Important
You must set the parameters to the exact values provided. Otherwise, the Cloud Assistant command may fail.
- Command Type: Select Shell.
- Command Content: Paste the following command. More Shell command examples: View the system configurations of ECS instances.
  - Cloud Assistant command for sgn8ia instances:
```
if acs-plugin-manager --list --local | grep gpu_grid_driver_install > /dev/null 2>&1
then
    acs-plugin-manager --remove --plugin gpu_grid_driver_install
fi
acs-plugin-manager --exec --plugin gpu_grid_driver_install
```
  - Cloud Assistant command for vgn6i-vws, vgn7i-vws, and sgn7i-vws instances:
```
if acs-plugin-manager --list --local | grep grid_driver_install > /dev/null 2>&1
then
    acs-plugin-manager --remove --plugin grid_driver_install
fi
acs-plugin-manager --exec --plugin grid_driver_install
```
- Timeout Period: The execution timeout in seconds. Cloud Assistant terminates the command if it exceeds this limit. Set to 600.
  
  Note
  The Timeout Period value must be an integer from 10 to 86,400 (24 hours).
Click Run to install the GRID driver by using the Cloud Assistant command.

View the execution result on the execution details page.

Important
If a different GRID driver version is already installed, Cloud Assistant automatically replaces it with the latest version.

A successful installation shows Execution Succeeded status, ExitCode 0, and a final log line of install gpu driver done.
Connect to the vGPU-accelerated instance.

Connect to a Linux instance by using Workbench.

Run the following command to verify the GRID driver installation.

nvidia-smi

If the output resembles the following, the GRID driver is successfully installed or upgraded.

[root@xxx ~]# nvidia-smi
Tue Jun  4 13:58:43 2024
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.256.03    Driver Version: 470.256.03    CUDA Version: 11.4   |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA A10-4Q       On   | 00000000:00:07.0 Off |                  N/A |
| N/A   N/A    P0    N/A /  N/A |    128MiB /  3932MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name              GPU Memory     |
|        ID   ID                                               Usage          |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

Test and verify

This section uses a GLX test program on an Ubuntu 18.04 instance to verify graphics acceleration.

Prepare the graphics test environment.
1. Run the following command to install x11vnc.
```
apt-get install x11vnc
```
2. Run the lspci | grep NVIDIA command to query the GPU BusID.
  
  In this example, the GPU BusID is 00:07.0.
3. Configure the X Server environment and restart the instance.
  1. Run the nvidia-xconfig --enable-all-gpus --separate-x-screens command.
  2. Edit the /etc/X11/xorg.conf file. In the Section "Device" block, add the BusID of the GPU. In this example, add BusID "PCI:0:7:0". Ensure that the BusID value in the file matches the GPU's PCI bus address.
```
Section "Device"
    Identifier    "Device0"
    Driver        "nvidia"
    VendorName    "NVIDIA Corporation"
    BoardName     "Tesla P4"
    BusID         "PCI:0:7:0"
EndSection
```
  3. Run the reboot command to restart the system.

Run the following command to install the GLX test program.

apt-get install mesa-utils

Run the startx command to start X Server.
- If the startx command is not found, run the apt-get install xinit command to install it.
- When you run startx, the message hostname: Name or service not known may appear. This message does not affect the X Server startup. You can run the hostname command to obtain the hostname, and then modify the /etc/hosts file. Change 127.0.0.1's corresponding hostname to your machine's hostname.
```
root@iZxxxZ:~# startx
hostname: Name or service not known
xauth: (stdin):1:  bad display name "iZxxxZ:1" in "add" command
```
Open a new SSH client terminal and run the following command to start x11vnc.
```
x11vnc -display :1
```
If the following output appears, x11vnc is running. Use a VNC client such as VNC Viewer to connect to the instance.
```
The VNC desktop is:      localhost:0
PORT=5900
```
Log on to the ECS console and add a security group rule to the security group of the instance to allow inbound access on TCP port 5900.

Add a security group rule.
On your local machine, use a VNC client such as VNC Viewer to connect to the instance at <public IP address of the instance>:5900 and access the KDE desktop.
Use the glxinfo command to view the driver's supported configurations.
1. Open a new SSH client terminal.
2. Run the export DISPLAY=:1 command.
3. Run the glxinfo -t command to list the configurations supported by the current GRID driver.
Follow these steps to test the GRID driver with the glxgears command.
1. On the KDE desktop, right-click the desktop and select Run Command.
2. Run the glxgears command to start the gears graphics test program.
  
  If a window that shows moving gears appears, the GRID driver is working as expected.

Elastic GPU Service:Install the GRID driver on a vGPU-accelerated instance (Linux)

Supported instances

Procedure

Test and verify

Related topics