GPU Installation

This appendix contains configuration rules for the supported GPU cards.

Supported GPUs and Server Firmware Requirements

The following table lists the minimum server firmware versions for the supported GPU cards.

GPU Card

PID

Type

Number of Supported GPUs*

Cisco IMC/BIOS Minimum Version Required

NVIDIA Tesla A10

UCSC-GPU-A10 or HX-GPU-A10=

Single-wide

5

4.2(2f)

NVIDIA Tesla A16

UCSC-GPU-A16=or HX-GPU-A16=

Double-wide

3

4.2(2f)

NVIDIA Tesla A30

UCSC-GPU-A30=or HX-GPU-A30=

Double-wide

3

4.2(2f)

NVIDIA Tesla A40

UCSC-GPU-A40 or HX-GPU-A40=

Double-wide

3

4.2(2f)

NVIDIA Tesla A100

UCSC-GPU-A100=

Double-wide

3

4.2(2f)

NVIDIA Tesla A100-80

UCSC-GPU-A100-80=

Double-wide

3

4.2(2f)

*The NVME server supports only 2 double-wide GPUs or 4 single-wide GPUs since it supports only two risers.

GPU Card Configuration Rules

Note the following rules when populating a server with GPU cards.

  • The UCSC-C240-M6SX and UCSC-C240-M6N servers support a "GPU Ready" config which presets the server to accept a GPU at a later date.

    This configuration sets the server with a with low-profile heatsink (UCSC-HSLP-M6) and GPU air duct (UCSC-ADGPU-240M6) so that only the GPU is required when you choose to expand the server's compute power. The GPU Ready config has the following considerations:

    • The server must follow the same temperature limits as a server with a GPU installed, even if the server does not currently have a GPU installed. By following the temperature limits even when the GPU is not present, you ensure correct operation when the GPU is installed later.

    • This configuration requires a low-profile heatsink and a GPU air duct. If you are ordering the GPU Ready configuration, you must select the GPU air duct PID to enable GPU ready configuration when ordering the server through the Cisco online ordering and configuration tool. Follow the additional rules displayed in the tool.

  • All GPU cards must be procured from Cisco because of a unique SBIOS ID that is required by CIMC and UCSM.

  • Do not mix different brands or models of GPU cards in the server.

  • NVIDIA Single Wide GPUs are supported:

    • The GPUs can be populated in Riser 1A slots 2 (x16) and 3 (x8), Riser 2 slots 5 (x16) and 6 (x8), and Riser 3C slot 7 (x16).

    • Each server can support a maximum of five of these GPUs.

  • NVIDIA Double Wide GPUs are supported:

    • This GPU requires x16 support.

    • This GPU can be populated in Riser 1A slot 2 (x16), Riser 2A slot 5 (x16), and Riser 3C slot 7 (x16).

    • Each server can support three of these GPUs.

  • GPUs are not supported in Riser 1B or Riser 3B. Riser 3B cannot mechanically accept a GPU.

  • The UCSC-C240M6-S and UCSC-C240M6-S servers support one full-height, full-length, double-wide GPU (PCIe slot 7 only) in Riser 3C.

  • The UCSC-C240-M6N and UCSC-C240-M6N servers do not support any GPU in slot 7 of Riser 3C.

  • UCSM managed servers are discoverable only if a PCIe VIC card is installed in slot 1 or slot 4 or an mLOM VIC card is installed in the mLOM slot. If you install double-width GPUs, they must be located in slots 2, 5, or 7. Therefore, if two GPUs are installed, UCSM managed servers are discoverable only if you install a VIC in slot 1, slot 4, or the mLOM slot. The server can support 2 PCIe VICs and 1 mLOM VIC along with 2 or 3 GPUs.

  • Use the UCS power calculator at the following link to determine the power needed based on your server configuration: http://ucspowercalc.cisco.com

Requirement For All GPUs: Memory-Mapped I/O Greater Than 4 GB

All supported GPU cards require enablement of the BIOS setting that allows greater than 4 GB of memory-mapped I/O (MMIO).

  • Standalone Server: If the server is used in standalone mode, this BIOS setting is enabled by default:

    Advanced > PCI Configuration > Memory Mapped I/O Above 4 GB [Enabled]

    If you need to change this setting, enter the BIOS Setup Utility by pressing F2 when prompted during bootup.

  • If the server is integrated with Cisco UCS Manager and is controlled by a service profile, this setting is enabled by default in the service profile when a GPU is present.

    To change this setting manually, use the following procedure.

Procedure

Step 1

Refer to the Cisco UCS Manager configuration guide (GUI or CLI) for your release for instructions on configuring service profiles:

Cisco UCS Manager Configuration Guides

Step 2

Refer to the chapter on Configuring Server-Related Policies > Configuring BIOS Settings.

Step 3

In the section of your profile for PCI Configuration BIOS Settings, set Memory Mapped IO Above 4GB Config to one of the following:

  • Disabled—Does not map 64-bit PCI devices to 64 GB or greater address space.

  • Enabled—Maps I/O of 64-bit PCI devices to 64 GB or greater address space.

  • Platform Default—The policy uses the value for this attribute contained in the BIOS defaults for the server. Use this only if you know that the server BIOS is set to use the default enabled setting for this item.

Step 4

Reboot the server.

Note

 

Cisco UCS Manager pushes BIOS configuration changes through a BIOS policy or default BIOS settings to the Cisco Integrated Management Controller (CIMC) buffer. These changes remain in the buffer and do not take effect until the server is rebooted.

Installing a Single-Wide GPU Card

Use the following procedure to install or replace an NVIDIA Tesla single-wide or dual GPUs.

With Cisco IMC version 4.2(1) and later, the server can support up to five single-wide GPUs.

Ordering PID

Description

UCSC-RIS1A-240M6

Riser 1A

UCSC-RIS2A-240M6

Riser 2A

UCSC-RIS3C-240M6

Riser 3C

Procedure

Step 1

Shut down and remove power from the server as described in Shutting Down and Removing Power From the Server.

Step 2

Slide the server out the front of the rack far enough so that you can remove the top cover. You might have to detach cables from the rear panel to provide clearance.

Caution

 
If you cannot safely view and access the component, remove the server from the rack.

Step 3

Remove the top cover from the server as described in Removing the Server Top Cover.

Step 4

Remove the single-wide GPU card that you are replacing:

  1. Use two hands to flip up and grasp the blue riser handle and the blue finger grip area on the front edge of the riser, and then lift straight up.

  2. On the bottom of the riser, push the release latch that holds the securing plate, and then swing the hinged securing plate open.

  3. Open the hinged card-tab retainer that secures the rear-panel tab of the card.

    Figure 1. PCIe Riser Card Securing Mechanisms

    1

    Release latch on hinged securing plate

    3

    Hinged card-tab retainer

    2

    Hinged securing plate

    -

  4. Disconnect the Y cable (UCSC-CBL-240M6) from the GPU ends (the black 8-pin connectors).

    If you are inserting another GPU, you can leave the opposite end of the cable (the white connector) attached. If you are not inserting another GPU, you can leave the cable in place or completely remove it.

  5. Pull evenly on both ends of the single-wide GPU card to remove it from the socket on the PCIe riser.

    If the riser has no card, remove the blanking panel from the rear opening of the riser.

Step 5

Install a new single-wide GPU card:

  1. With the hinged card-tab retainer open, align the new single-wide GPU card with the empty socket on the PCIe riser.

  2. Push down evenly on both ends of the card until it is fully seated in the socket.

  3. Ensure that the card’s rear panel tab sits flat against the riser rear-panel opening and then close the hinged card-tab retainer over the card’s rear-panel tab.

  4. Swing the hinged securing plate closed on the bottom of the riser. Ensure that the clip on the plate clicks into the locked position.

  5. Position the PCIe riser over its socket on the motherboard and over the chassis alignment channels.

  6. Carefully push down on both ends of the PCIe riser to fully engage its connector with the sockets on the motherboard.

  7. Attach the Y cable to the new GPU by connecting the white end to the motherboard and one of the black ends to each GPU.

    Note

     

    If the server has only one GPU in the riser, one of the black GPU cables will be unused. You can let it dangle in place.

Step 6

Replace the top cover to the server.

Step 7

Replace the server in the rack, replace power and network cables, and then fully power on the server by pressing the Power button.

Step 8

Optional: Continue with Installing Drivers to Support the GPU Cards.

Installing a Double-Wide GPU Card

Use the following procedure to install or replace a NVIDIA Tesla Double-Wide GPU.

With Cisco IMC version 4.2(1) and later, the server can support up to three Double-Wide GPUs.

Server

Component

Recommended T Ambient Spec, Normal

All C240 24 SFF

PCIe

35 C

Rear HDD

30 C

GPU A100

GPU A10

C240 12 LFF

Rear HDD

30 C

GPU A100

GPU A10

28 C
The NVIDIA GPU card might be shipped with two power cables: a straight cable and a Y-cable. The straight cable is used for connecting power to the GPU card in this server; do not use the Y-cable, which is used for connecting the GPU card in external devices only.

The supported NVIDIA GPU requires a C240 M5 NVIDIA Cable (UCS-P100CBL-240M5).

Procedure

Step 1

Shut down and remove power from the server as described in Shutting Down and Removing Power From the Server.

Step 2

Slide the server out the front of the rack far enough so that you can remove the top cover. You might have to detach cables from the rear panel to provide clearance.

Caution

 
If you cannot safely view and access the component, remove the server from the rack.

Step 3

Remove the top cover from the server as described in Removing the Server Top Cover.

Step 4

Remove an existing GPU card:

  1. Disconnect any existing cable from the GPU card.

  2. Use two hands to grasp the metal bracket of the PCIe riser and lift straight up to disengage its connector from the socket on the motherboard. Set the riser on an antistatic surface.

  3. On the bottom of the riser, press down on the clip that holds the securing plate.

  4. Swing open the hinged securing plate to provide access.

  5. Open the hinged plastic retainer that secures the rear-panel tab of the card.

  6. Disconnect the GPU card's power cable from the power connector on the PCIe riser.

  7. Pull evenly on both ends of the GPU card to remove it from the socket on the PCIe riser.

Figure 2. PCIe Riser Card Securing Mechanisms

1

Release latch on hinged securing plate

3

Hinged card-tab retainer

2

Hinged securing plate

-

Step 5

Install a new GPU card:

Note

 

Observe the configuration rules for this server, as described in GPU Card Configuration Rules.

  1. Align the GPU card with the socket on the riser, and then gently push the card’s edge connector into the socket. Press evenly on both corners of the card to avoid damaging the connector.

  2. Connect the GPU power cable. The straight power cable connectors are color-coded. Connect the cable's black connector into the black connector on the GPU card and the cable's white connector into the white GPU POWER connector on the PCIe riser.

    Caution

     

    Do not reverse the straight power cable. Connect the black connector on the cable to the black connector on the GPU card. Connect the white connector on the cable to the white connector on the PCIe riser.

  3. Close the card-tab retainer over the end of the card.

  4. Swing the hinged securing plate closed on the bottom of the riser. Ensure that the clip on the plate clicks into the locked position.

  5. Position the PCIe riser over its socket on the motherboard and over the chassis alignment channels.

  6. Carefully push down on both ends of the PCIe riser to fully engage its connector with the sockets on the motherboard.

    At the same time, align the GPU front support bracket (on the front end of the GPU card) with the securing latch that is on the server's air baffle.

Step 6

Insert the GPU front support bracket into the latch that is on the air baffle:

  1. Pinch the latch release tab and hinge the latch toward the front of the server.

  2. Hinge the latch back down so that its lip closes over the edge of the GPU front support bracket.

  3. Ensure that the latch release tab clicks and locks the latch in place.

Step 7

Replace the top cover to the server.

Step 8

Replace the server in the rack, replace power and network cables, and then fully power on the server by pressing the Power button.

Step 9

Optional: Continue with Installing Drivers to Support the GPU Cards.

Enabling GPU Display Ports

Enabling GPU Display Ports

By default, system output (including from GPUs) is sent to the KVM. However, some models of NVIDIA GPU, such as the A40 and L40, have the ability to send output to the GPU display port. Where supported, this feature might be disabled by default depending on which GPU you are using.

You can selectively control enabling or disabling the GPU display mode by using a free tool (called displaymodeselector) which available through a shared Cisco Box folder. This tool enables sending output to the GPU display ports instead of the KVM.

See the following topics:

GPU Display Port Considerations

When enabling or disabling the GPU display ports on NVIDIA GPUs, be aware of the following guidelines and considerations.

  • By default, display port functionality is disabled on NVIDIA A40 and L40 GPUs. With the display port functionality disabled, standard server output is sent to the KVM. If you want to use display ports on these GPUs, you must explicitly enable them through the displaymodeselector tool.

  • When display port functionality is disabled, both CUDA compute drivers and GRID graphics drivers work in this mode.

  • When the display mode is enabled state, the NVIDIA Accelerated Linux Graphics Driver must be used.

    • https://cisco.box.com/s/wf8flcrfvobd8i8vgap1xstzt8ls3x17

    • vGPU drivers cannot be used.

    • When the display mode is in enabled state, an A40 GPU can never be the primary graphics boot device. You must keep any A40 GPU as a secondary and use either an on-board VGA (in server) or another NVIDIA GPU as the primary boot device to drive displays.

  • After changing the display mode state, do not run NVQUAL.

  • When running displaymodeselector on a Windows server, additional configuration steps are required.

    Use the documentation at:

    https://cisco.box.com/s/wf8flcrfvobd8i8vgap1xstzt8ls3x17

  • When running displaymodeselector on a Linux server, you must also create the xorg.conf file.

  • Anytime you change the state of GPU display ports through displaymodeselector, or edit or change the xorg.conf file, you must reboot the server.

Understanding the displaymodeselector Tool

The displaymodeselector tool is freeware available to download and install from the following shared Cisco Box:

https://cisco.box.com/s/26yx4j82hbxce6r80tbhmnkuyvnivj9q

The tool is supported on both Linux and Windows. You run the tool by issuing the following command:

displaymodeselector --gpumode <mode_name> [-i num]

Note

If you are running this tool on Windows, additional steps are required. See the documentation in the following shared Cisco Box:

https://cisco.box.com/s/wf8flcrfvobd8i8vgap1xstzt8ls3x17

The tool supports the following display options.

Display Option

Used For

Physical Display Ports Enabled with 256MB BAR1

Standard workstation deployments that contain physically attached displays.

Physical Display Ports Enabled with 8GB BAR1

Specific use cases, such as broadcast, virtual production, and location-based entertainment deployments, which require physical display ports as well as the support of additional performance optimizations through an 8 BAR1 such as the NVIDIA Rivermax software.

Physical Display Ports Disabled

Running NVIDIA virtual GPU (vGPU) software or compute use cases where no physically attached displays are required.

Enabling or Disabling GPU Display Ports

Use the gpumodeselector tool to set the state of display port functionality.

Before you begin

Review the GPU Display Port Considerations.

Procedure

Step 1

After starting the tool, execute the command:

displaymodeselector --gpumode <mode_name> [-i num]

where:

mode-name

Is one of the following mandatory arguments:

  • physical_display_enabled_256MB_bar1

  • physical_display_enabled_8GB_bar1

  • physical_display_disabled

    Note

     

    For Linux servers, you can also disable the GPU display ports by either deleting the xconf.org file or making it an empty (zero-byte) file.

-i num

Is an optional argument that specifies the GPU index number.

Step 2

Verify the register specification.

Note

 

By default, BIOS assigns the BAR1 registers. You can use lspci to verify which registers are in use. Consult the following table to determine which registers to use with the mode-name argument.

Software Specification

Description

SR-IOV Support

32 VF (virtual functions)

BAR address (physical functions)

  • BAR0: 16 MiB (mebibyte)

  • BAR1: 64 GiB (gibibyte). Display Off mode, which is the default.

  • BAR1: 8 GiB (gibibyte). Display mode on, 8 GB (gigabyte) BAR1 mode.

  • BAR1: 256 MiB (mibibyte). Display On mode,256 MB (megabyte) BAR1 mode.

  • BAR3: 32 MiB (mebibytes)

BAR address (virtual functions)

With Display Off modes (default):

  • BAR0: 8 MiB (32 VF * 256 KiB, or kebibyte)

  • BAR1: 64 GiB, 64 bit (32 VF * 2 GiB)

  • BAR3: 1 GiB, 64 bit (32 VF * 32 MiB)

  • BAR1: 256 MiB (mibibyte). Display mode on,256 MB (megabyte) BAR1 mode.

  • BAR3: 32 MiB (mebibyte)

With display mode on, VF BAR sizes are not applicable to Display On modes.

Step 3

Choose one of the following, depending the OS running on the server:

  • For Windows servers, after setting the display mode, complete the configuration as described in the documentation here:

    https://cisco.box.com/s/wf8flcrfvobd8i8vgap1xstzt8ls3x17

  • For Linux servers, after setting the display mode, create the xconf.org file.

    Go to Creating the xconf File for Linux Servers.

Step 4

You must reboot the server to complete configuring the display ports after configuring the server with OS specific settings.

Creating the xconf File for Linux Servers

By default, the xconf.org file does not exist, so you will need to create it. Use this task to create the file and add the required parameters.


Note

This task is applicable only to servers running Linux.

Procedure

Step 1

Execute nvidia-xconfig to create the file with default parameters.

When the command completes, you will find the default xconf.org file in /etc/X11/ location.

Step 2

Using vi, emacs, or another ASCII text editor, view the newly created file and verify that /etc/X11/xconf.org contains the following stanzas.

Caution

 
Multiple displays might be listed. Multiple displays are not supported.

Step 3

Reboot the server whenever you change the xconf.org file.

Step 4

After the server reboots, check /var/log/Xorg.0.log to see the GPU entry (an A40 in the following example) and the supported display.

Reverting Display Mode to KVM

At any time, you can revert the display from the GPU display ports to the default KVM. Use the following task to return output signals to the KVM. The KVM must be connected to see the output.

Procedure

Step 1

Choose the appropriate option.

  • Using vi, emacs, or another ASCII-text editor, delete all contents of xconf.org, leaving it a zero-byte file.

  • Completely delete (rm) the xconf.org file.

Step 2

Reboot the server to activate the changes and send the output to the KVM.

Installing GRID Licenses From the NVIDIA Licensing Portal to the License Server

Accessing the GRID License Server Management Interface

Open a web browser on the License Server host and access the URL http://localhost:8080/licserver.

If you configured the License Server host’s firewall to permit remote access to the License Server, the management interface is accessible from remote machines at the URL http://hostname:8080/licserver

Reading Your License Server’s MAC Address

Your License Server’s Ethernet MAC address is used as an identifier when registering the License Server with NVIDIA’s Licensing Portal.

Procedure

Step 1

Access the GRID License Server Management Interface in a browser.

Step 2

In the left-side License Server panel, select Configuration.

The License Server Configuration panel opens. Next to Server host ID, a pull-down menu lists the possible Ethernet MAC addresses.

Step 3

Select your License Server’s MAC address from the Server host ID pull-down.

Note

 

It is important to use the same Ethernet ID consistently to identify the server when generating licenses on NVIDIA’s Licensing Portal. NVIDIA recommends that you select one entry for a primary, non-removable Ethernet interface on the platform.

Installing Licenses From the Licensing Portal

Procedure

Step 1

Access the GRID License Server Management Interface in a browser.

Step 2

In the left-side License Server panel, select Configuration.

The License Server Configuration panel opens.

Step 3

Use the License Server Configuration menu to install the .bin file that you generated earlier.

  1. Click Choose File.

  2. Browse to the license .bin file that you want to install and click Open.

  3. Click Upload.

    The license file is installed on your License Server. When installation is complete, you see the confirmation message, “Successfully applied license file to license server.”

Viewing Available GRID Licenses

Use the following procedure to view which licenses are installed and available, along with their properties.

Procedure

Step 1

Access the GRID License Server Management Interface in a browser.

Step 2

In the left-side License Server panel, select Licensed Feature Usage.

Step 3

Click on a feature in the Feature column to see detailed information about the current usage of that feature.

Viewing Current License Usage

Use the following procedure to view information about which licenses are currently in-use and borrowed from the server.

Procedure

Step 1

Access the GRID License Server Management Interface in a browser.

Step 2

In the left-side License Server panel, select Licensed Clients.

Step 3

To view detailed information about a single licensed client, click on its Client ID in the list.

Managing GRID Licenses

Features that require GRID licensing run at reduced capability until a GRID license is acquired.

Acquiring a GRID License on Windows

Procedure

Step 1

Open the NVIDIA Control Panel using one of the following methods:

  • Right-click on the Windows desktop and select NVIDIA Control Panel from the menu.

  • Open Windows Control Panel and double-click the NVIDIA Control Panel icon.

Step 2

In the NVIDIA Control Panel left-pane under Licensing, select Manage License.

The Manage License task pane opens and shows the current license edition being used. The GRID software automatically selects the license edition based on the features that you are using. The default is Tesla (unlicensed).

Step 3

If you want to acquire a license for GRID Virtual Workstation, under License Edition, select GRID Virtual Workstation.

Step 4

In the License Server field, enter the address of your local GRID License Server. The address can be a domain name or an IP address.

Step 5

In the Port Number field, enter your port number of leave it set to the default used by the server, which is 7070.

Step 6

Select Apply.

The system requests the appropriate license edition from your configured License Server. After a license is successfully acquired, the features of that license edition are enabled.

Note

 

After you configure licensing settings in the NVIDIA Control Panel, the settings persist across reboots.

Acquiring a GRID License on Linux

Procedure

Step 1

Edit the configuration file /etc/nvidia/gridd.conf:

sudo vi /etc/nvidia/gridd.conf

Step 2

Edit the ServerUrl line with the address of your local GRID License Server.

The address can be a domain name or an IP address. See the example file below.

Step 3

Append the port number (default 7070) to the end of the address with a colon. See the example file below.

Step 4

Edit the FeatureType line with the integer for the license type. See the example file below.

  • GRID vGPU = 1

  • GRID Virtual Workstation = 2

Step 5

Restart the nvidia-gridd service.

sudo service nvidia-gridd restart

The service automatically acquires the license edition that you specified in the FeatureType line. You can confirm this in /var/log/messages.

Note

 

After you configure licensing settings in the NVIDIA Control Panel, the settings persist across reboots.

Sample configuration file:

# /etc/nvidia/gridd.conf - Configuration file for NVIDIA Grid Daemon
# Description: Set License Server URL
# Data type: string
# Format: "<address>:<port>" 
ServerUrl=10.31.20.45:7070

# Description: Set Feature to be enabled
# Data type: integer
# Possible values:
# 1 => for GRID vGPU 
# 2 => for GRID Virtual Workstation
FeatureType=2

Using gpumodeswitch

The command line utility gpumodeswitch can be run in the following environments:

  • Windows 64-bit command prompt (requires administrator permissions)

  • Linux 32/64-bit shell (including Citrix XenServer dom0) (requires root permissions)


Note

Consult NVIDIA product release notes for the latest information on compatibility with compute and graphic modes.

The gpumodeswitch utility supports the following commands:

  • --listgpumodes

    Writes information to a log file named listgpumodes.txt in the current working directory.

  • --gpumode graphics

    Switches to graphics mode. Switches mode of all supported GPUs in the server unless you specify otherwise when prompted.

  • --gpumode compute

    Switches to compute mode. Switches mode of all supported GPUs in the server unless you specify otherwise when prompted.


Note

After you switch GPU mode, reboot the server to ensure that the modified resources of the GPU are correctly accounted for by any OS or hypervisor running on the server.

Installing Drivers to Support the GPU Cards

After you install the hardware, you must update to the correct level of server BIOS and then install GPU drivers and other software in this order:

  1. Update the server BIOS.

  2. Update the GPU drivers.

1. Updating the Server BIOS

Install the latest Cisco UCS C240 M4 server BIOS by using the Host Upgrade Utility for the Cisco UCS C240 M4 server.


Note

You must do this procedure before you update the NVIDIA drivers.
Procedure

Step 1

Navigate to the following URL: http://www.cisco.com/cisco/software/navigator.html.

Step 2

Click Servers–Unified Computing in the middle column.

Step 3

Click Cisco UCS C-Series Rack-Mount Standalone Server Software in the right-hand column.

Step 4

Click the name of your model of server in the right-hand column.

Step 5

Click Unified Computing System (UCS) Server Firmware.

Step 6

Click the release number.

Step 7

Click Download Now to download the ucs-server platform-huu-version_number.iso file.

Step 8

Verify the information on the next page, and then click Proceed With Download.

Step 9

Continue through the subsequent screens to accept the license agreement and browse to a location where you want to save the file.

Step 10

Use the Host Upgrade Utility to update the server BIOS.

The user guides for the Host Upgrade Utility are at Utility User Guides.

2. Updating the GPU Card Drivers

After you update the server BIOS, you can install GPU drivers to your hypervisor virtual machine.

Procedure

Step 1

Install your hypervisor software on a computer. Refer to your hypervisor documentation for the installation instructions.

Step 2

Create a virtual machine in your hypervisor. Refer to your hypervisor documentation for instructions.

Step 3

Install the GPU drivers to the virtual machine. Download the drivers from either:

Step 4

Restart the server.

Step 5

Check that the virtual machine is able to recognize the GPU card. In Windows, use the Device Manager and look under Display Adapters.