GPU passthrough and vGPU settings

This section uses a NVIDIA Tesla M60 GPU as an example to describe GPU and vGPU settings. For information about server configurations, see "Server configurations." For information about GPU passthrough settings, see "GPU passthrough." For information about vGPU settings, see "vGPU settings."

Server configurations

1. Power off the server.

2. Power on the server, and press F9 as instructed at server startup to access the BIOS setup utility. Then, access the Advanced Options > Video Options screen, and set the embedded video as the primary and the optional video as the secondary.

Figure-1 Setting the embedded video as the primary and the optional video as the secondary

 

3. Exit the BIOS and restart the server.

4. Install CVK and CVM on the server.

5. Log in to the CAS management platform, access the Resources > Host Pool Name > Cluster Name > Host Name > Advanced page, and then enable IOMMU.

6. Restart the server.

GPU passthrough

1. Log in to the CAS management platform and access the Services > iRS page.

2. Click Add iRS Service.

3. Specify the service name, select the target cluster, and set the resource type to GPU.

4. Click Next and then select the target GPU devices on the hosts in the cluster.

5. Click Next and then directly click Finish.

Cloud host settings are not required in this step.

6. Create a GPU cloud host as described in the quick start.

7. Install a driver for the CPU display card after transparent transmission. Upload the corresponding GPU device driver into the cloud host, and then install the display card driver program in the cloud host. After installation, restart the cloud host.

vGPU settings

About NVIDIA vGPUs

vGPUs have a fixed amount of frame buffer, number of supported display heads, and maximum resolutions. They are grouped into the following series based on workload:

• Q series—For designers and premier users.

• B series—For premier users.

• A series—For virtual application users.

The number of vGPUs that can be created on a physical GPU is limited according to the following principles:

• vGPUs are created by frame buffer. vGPUs created on the same physical GPU must have the same frame buffer.

• You can create vGPUs on a physical GPU with the same frame buffer.

• You can use different physical GPUs on the same graphics card to create vGPUs of different types.

For example, a Tesla M60 graphics card has two physical GPUs and 16 GB buffer in total (8 GB for each GPU). You can create vGPUs with a frame buffer size of 0.5 GB, 1 GB, 2 GB, 4 GB, or 8 GB. Table-1 displays vGPU types supported on a Tesla M60.

Table-1 vGPU types supported on a Tesla M60

vGPU type	Frame buffer (MB)	Maximum display heads	Maximum display resolution	Maximum vGPUs per physical GPU	Maximum vGPUs per graphics card
M60-8Q	8192	4	4096×2160	1	2
M60-4Q	4096	4	4096×2160	2	4
M60-2Q	2048	4	4096×2160	4	8
M60-1Q	1024	2	4096×2160	8	16
M60-0Q	512	2	2560×1600	16	32
M60-2B	2048	2	4096×2160	4	8
M60-1B	1024	4	2560×1600	8	16
M60-0B	512	2	2560×1600	16	32
M60-8A	8192	1	1280×1024	1	2
M60-4A	4096	1	1280×1024	2	4
M60-2A	2048	1	1280×1024	4	8
M60-1A	1024	1	1280×1024	8	16

 

CAS does not support creating 512 MB vGPUs, such as M60-0Q and M60-0B. For more information about NVIDIA GPUs applicable to virtualization and their supported vGPUs, see Virtual GPU Software User Guide at https://docs.nvidia.com/grid/latest/grid-vgpu-user-guide/index.html.

 

About vGPU licensing

NVIDIA GRID vGPU is a licensed software product. At cloud host startup, the cloud host must borrow a license from the NVIDIA vGPU License server through the Internet to activate vGPU functions. At cloud host shutdown, the cloud host returns the license to the license server.

Figure-2 NVIDIA GRID vGPU licensing structure

 

A NVIDIA Tesla GPU provides the following NVIDIA GRID licensed products:

• Virtual Workstation.

• Virtual PC.

• Virtual Application.

Table-2 displays available GRID license types.

Table-2 GRID license types

GRID License type	GRID functions	Supported vGPUs
GRID Virtual Application	Provides PC-level applications.	A-series vGPUs
GRID Virtual PC	Provides virtual desktop services to Windows, Web browsers, and high-definition video applications.	B-series vGPUs
GRID Virtual Workstation	Serves medium- to high-end workstation users that require remote access to professional graphic applications. It uses GPU passthrough for graphics calculation.	Q-series and B-series vGPUs

 

Prerequisites

Obtain the installation package for Virtual GPU License Manager, the gpumodeswitch tool, and the GPU driver from the official website of NVIDIA. For more information, see "About NVIDIA vGPUs."

NVIDIA Tesla M60 and M6 support both the compute mode and the graphics mode. To create vGPU devices by using a NVIDIA Tesla M60 or M6 card, use the gpumodeswitch tool to switch the mode of the card to graphics. For more information, see gpumodeswitch User Guide at https://docs.nvidia.com/grid/6.0/grid-gpumodeswitch-user-guide/index.html.

(Optional.) Deploy the NVIDIA License Server

To use full NVIDIA vGPU functions, you must deploy the NVIDIA License Server and register a license. For more information, see "About vGPU licensing."

 

Hardware requirements

• You can install the license server on a physical server or a cloud host.

• As a best practice, make sure the host has a minimum of two CPU cores and 4 GB memory.

• To support up to 150,000 licensed clients, a minimum of four CPU cores and 16 GB memory are required.

Operating system requirements

As a best practice, make sure the operating system is of the English edition. If you use a non-English operating system, set the browser language to English. For more information, see "Set the browser language."

The recommended operating systems of NVIDIA include:

• Windows:

• Windows 10 64-bit

• Windows Server 2016

• Windows Server 2012 R2

• Linux:

• Red Hat Enterprise Linux 7.3 64-bit

• CentOS 7.3 64-bit

To use a Linux operating system, you must set up the graphical desktop environment.

Operating environment requirements

• Use 32-bit JRE of version 1.8 or higher. Install JRE before installing the license server.

• To use a Windows OS, make sure the version of NET Framework is 4.5 or higher.

• For a Windows license server, you do not need to install Apache Tomcat because the installation package of the server incorporates the Apache Tomcat package. For a Linux license server, you must install Apache Tomcat before installing the license server. Make sure Apache Tomcat is of version 7x or 8x.

• Use Firefox 17, Chrome 27, IE 9, or a browser with a higher version.

Server requirements

• Make sure the server has a fixed IP address.

• Make sure the server has a fixed MAC address, which will be used as the unique identifier of the server during license generation on the license server and NVIDIA portal.

• Make sure the system date and time are configured correctly.

Network port and access methods

The license server uses port 7070 to provide registration services. Make sure TCP port 7070 is open on the firewall.

You can use either of the following methods to access the license server:

• Visit http://localhost:8080/licserver on the host on which the license server resides.

• Visit http://serverip:8080/licserver on a PC that can reach the license server. The serverip argument represents the IP address of the host on which the license server resides.

Procedure

1. Log in to the CAS management platform and create a cloud host for NVIDIA License Server deployment.

2. Install an operating system for the cloud host as required.

3. Install NVIDIA License Manager. For more information, see installing the NVIDIA vGPU Software License Server in License Server User Guide at https://docs.nvidia.com/grid/6.0/grid-license-server-user-guide/index.html.

4. Configure the license server. For more information, see Manager Licenses on the NVIDIA vGPU Software License Server in License Server User Guide.

Create a vGPU device

1. Log in to the CAS management platform and access the Resources > Host Pool Name > Cluster Name > Host Name or Resources > Host Pool Name > Host Name page.

2. Click the GPUs tab.

3. Click the  icon for the target GPU.

4. Select a vGPU name and click OK.

Add an iRS service

1. Log in to the CAS management platform and access the Services > iRS page.

2. Click Add iRS Service.

3. Specify the service name, select the target cluster, and set the resource type to vGPU.

4. Click Next and then select the target vGPU devices on the hosts in the cluster.

5. Click Next and then directly click Finish.

Cloud host settings are not required in the last step.

Create a vGPU cloud host and install the display driver

1. Create a vGPU cloud host as described in the quick start.

2. Install the NVIDIA display driver in the cloud host. Upload the corresponding GPU device driver into the cloud host, and then install the display card driver program in the cloud host. After installation, restart the cloud host.

Configure vGPU licenses

1. Access the operating system of the cloud host, right-click on the desktop, and select NVIDIA Control Panel from the dialog box that opens.

2. Select Licensing > Manage License, specify the IP address of the license server's host, set the port number to 7070, and then click Apply.

Set the browser language

1. Access IE and open Internet Options.

2. On the General tab, click Languages.

3. Click Add and add English to the list.

4. Change the order of English to the first, click OK to save the change, and then restart the browser.