1. vGPU
Note : Only NVIDIA's graphic drivers are compatible. Other solutions such as AMD's MxGPU are not compatible.
-
Supported Hardware Models for vGPU
- Partner's compatible HW list of NVIDIA vGPU: vgpu-certified-servers.
- Partner's compatible HW list of NVIDIA vGPU: vgpu-certified-servers.
-
Graphics Cards Compatible with vGPU
- vGPU supported graphics card list : gpus-supported-by-vgpu
- For instance, the NVIDIA RTX A5000 requires the Displayless mode to be configured properly.
Switching GPU Modes for Optimal Performance
-
Some GPUs support various display modes and are configurable for optimal performance with NVIDIA vGPU software. The following table indicates the default modes for GPUs as shipped from the factory:
-
GPU Mode as Supplied from the Factory NVIDIA A40 Displayless NVIDIA RTX A5000 Display enabled NVIDIA RTX A6000 Display enabled -
Utilize the NVIDIA Display Mode Selector Tool to make necessary adjustments.
-
- vGPU supported graphics card list : gpus-supported-by-vgpu
-
NVIDIA Virtual GPU Licensing
- vGPU function requires paid license. A 90-day trial is available for evaluating the vGPU's capabilities. But after 90 days vgpu operation will stop.
- vGPU function requires paid license. A 90-day trial is available for evaluating the vGPU's capabilities. But after 90 days vgpu operation will stop.
-
Recommended Software Environment
- VMware vSphere ESXi 7.0 Update 2 Enterprise Edition (a 60-day trial is available)
- Ensure compatibility with ESXi 7.0 Update 2 and subsequent updates.
- Note: NVIDIA Virtual Compute Server (vCS) is not supported on VMware vSphere ESXi. C-series GPU updates are not available. Instead, vCS is supported with NVIDIA AI Enterprise.
-
BIOS Settings for vGPU
- Ensure that VT-d/IOMMU is enabled in the BIOS settings.
- For servers with NVIDIA Ampere architecture GPUs, also enable SR-IOV in the BIOS advanced options.
2. GPU Passthrough
Prerequisites:
- Ensure your NVIDIA GPU is listed on the VMware Compatibility Guide.
- Update to the latest version of VMware ESXi.
Configuration Steps:
-
-
Identify the GPU:
- Connect to the ESXi host via SSH and use the
lspci
command to ensure the GPU is recognized by the system.
ESXI GUI
- Connect to the ESXi host via SSH and use the
-
Enable Passthrough:
- Navigate to the "Configure" tab in the ESXi GUI, select "Hardware," choose "PCI Devices," and enable passthrough for the GPU.
- Navigate to the "Configure" tab in the ESXi GUI, select "Hardware," choose "PCI Devices," and enable passthrough for the GPU.
-
VM Settings:
- Adjust the virtual machine settings to accommodate the GPU passthrough. This may include setting
hypervisor.cpuid.v0 = FALSE
if necessary, but note that it could affect the number of vCPUs available to the VM.
- Adjust the virtual machine settings to accommodate the GPU passthrough. This may include setting
-
Driver Installation:
- Install the correct NVIDIA driver within the virtual machine's operating system.
-
BIOS Settings:
- Ensure that VT-d/IOMMU is enabled in the BIOS, and SR-IOV is enabled if you're using GPUs based on the NVIDIA Ampere architecture.
-
Verify the Passthrough:
- After the VM is configured and the drivers are installed, verify that the GPU is recognized and functioning correctly
-