How to ensure the pass through vGPU being added to a VM is a __shared__ (not dedicated) vGPU pass through device

[cjoshi]

Hi,

I am adding a nvidia grid vGPU to a VM using the following code:

---- cut ----

var vGPUProfile = 'grid_m10-1b';

var deviceConfigSpec = new VcVirtualDeviceConfigSpec();

deviceConfigSpec.operation = VcVirtualDeviceConfigSpecOperation.add;

deviceConfigSpec.device = new VcVirtualPCIPassthrough;

deviceConfigSpec.device.deviceInfo = new VcDescription;

deviceConfigSpec.device.deviceInfo.summary = '';

deviceConfigSpec.device.deviceInfo.label = 'New PCI device';

deviceConfigSpec.device.backing = new VcVirtualPCIPassthroughVmiopBackingInfo;

deviceConfigSpec.device.backing.vgpu = vGPUProfile;

var deviceConfigSpecs = new Array();

deviceConfigSpecs[0] = deviceConfigSpec;

var configSpec = new VcVirtualMachineConfigSpec();

configSpec.changeVersion = vm.config.changeVersion;

configSpec.memoryReservationLockedToMax = true;

configSpec.deviceChange = deviceConfigSpecs;

task = vm.reconfigVM_Task(configSpec);

---- cut ----

In vRA the above workflow is triggered via EBS, the VM gets configured with the vGPU profiles. But fails to power-on with an error "exceeds max number of vGPU instances allowed".  I believe the error message is misleading and the VM is failing to get powered on because the vGPU is getting added in a dedicated mode instead of shared mode.

Question is how do I ensure the vGPU being added is forced to be as a Shared-vGPU device? Or am I missing something?

Thanks.

Regards Shekhar

[cjoshi · Accepted solution]

I am replying to this question. Since after a bit of digging around. I realized that the Cluster did not have enough vGPU capacity to power on a VM. TO essential the below mentioned code works as expected.