32 lines
1.4 KiB
YAML
32 lines
1.4 KiB
YAML
---
|
|
features:
|
|
- |
|
|
When using XenAPI driver for XenServer, we can support booting instances
|
|
with a vGPU attached to get better graphics processing capability.
|
|
In order to use this feature, the operators should specify the enabled
|
|
vGPU types in the nova compute configuration file with the configuration
|
|
option - ``[devices]/enabled_vgpu_types``. Only the enabled vGPU types
|
|
can be used by instances.
|
|
|
|
XenServer automatically detects and groups together identical physical
|
|
GPUs. Although the physical GPUs may support multiple vGPU types, at
|
|
the moment nova only supports a single vGPU type for each compute node.
|
|
The operators can run the following CLI commands in XenServer to get
|
|
the available vGPU types if the host supports vGPU::
|
|
|
|
xe vgpu-type-list
|
|
|
|
The values of ``model-name ( RO):`` from the output of the above commands
|
|
are the vGPU type names which you can choose from to set the nova
|
|
configure - ``[devices]/enabled_vgpu_types``. Please choose only one
|
|
vGPU type to be enabled.
|
|
|
|
The operators should specify a vGPU resource in the flavor's extra_specs::
|
|
|
|
nova flavor-key <flavor-id> set resources:VGPU=1
|
|
|
|
Then users can use the flavor to boot instances with a vGPU attached.
|
|
At the moment, XenServer doesn't support multiple vGPUs for a single
|
|
instance, so ``resources:VGPU`` in the flavor's extra_specs should
|
|
always be ``1``.
|