I see, thanks Suresh for this information. So just so I understand it 
correctly, at the moment, there is no support for specifying the no. of GPU 
that will be used by a compute offering, correct? Will this be supported in the 
future releases?

Thanks,
Irvin

> On Aug 13, 2021, at 2:37 PM, Suresh Anaparti <[email protected]> 
> wrote:
> 
> There is no provision to specify GPU count through service offering.
> 
> Regards,
> Suresh
> 
> On 13/08/21, 11:52 AM, "SVI" <[email protected]> wrote:
> 
>    Sorry, I’m not sure I follow. So right now, my XCP-NG hypervisor detects 8 
> GPU cards (Tesla K80). I added a compute offering with passthrough as 
> vgpuType and I was able to successfully create an instance and passthrough 1 
> gpu. I was able to validate this as well in the guest VM. I was actually 
> hoping that in the createServiceOffering API that there’s a GPU count (just 
> like cpu count) so whenever I use that compute offering it will passthrough 
> that no. of GPU count in the guest VM.
> 
>    Thanks,
>    Irvin
> 
> 
> 
> 
>> On Aug 13, 2021, at 1:48 PM, Suresh Anaparti <[email protected]> 
>> wrote:
>> 
>> May be, you can try this out-of-band. Check / test with the following - stop 
>> VM, add GPUs and start VM from the hypervisor.
>> 
>> Regards,
>> Suresh
>> 
>> On 13/08/21, 10:31 AM, "SVI" <[email protected]> wrote:
>> 
>>   Hi Suresh,
>> 
>>   I’m using Cloudstack 4.15.1 and XCP-NG 8.2.0 as the hypervisor host. Does 
>> this means that multiple GPU passthrough to 1 guest VM is not supported at 
>> the moment? What could be my other options here? I have a requirement for 
>> multiple GPUs in 1 VM that’s why I’m looking for this solution.
>> 
>> 
>>   Thanks,
>>   Irvin
>> 
>> 
>> 
>> 
>>> On Aug 13, 2021, at 12:31 PM, Suresh Anaparti 
>>> <[email protected]> wrote:
>>> 
>>> Hi Irvin,
>>> 
>>> The compute service offering in CloudStack currently supports 1 GPU 
>>> specification (using createServiceOffering  API through 
>>> 'serviceofferingdetails' parameters, with the keys pciDevice and vgpuType 
>>> for GPU card and vGPU type respectively), and so 1 card is created for a VM.
>>> 
>>> You can find some documentation here:
>>> 
>>> - 
>>> http://docs.cloudstack.apache.org/en/latest/adminguide/virtual_machines.html#assigning-gpu-vgpu-to-guest-vms
>>> 
>>> - 
>>> http://docs.cloudstack.apache.org/en/latest/adminguide/service_offerings.html#creating-a-new-compute-offering
>>> 
>>> May I know the CloudStack release and Hypervisor version you are testing.
>>> 
>>> Regards,
>>> Suresh
>>> 
>>> On 13/08/21, 7:03 AM, "SVI" <[email protected]> wrote:
>>> 
>>>  Hi,
>>> 
>>>  I was able to create a GPU passthrough offering but at the moment it’s 1 
>>> VM = 1 GPU passthrough. I have 8 GPUs in my hypervisor and technically I 
>>> can create 8 instance with 1 GPU passthrough each. What I want to do now is 
>>> to create at least 2 GPU to be passthrough 1 guest VM. I can’t see any 
>>> configuration or document to do that as of the moment, is there a hidden 
>>> API (I checked already the service offering API and doesn’t say much) that 
>>> is out there to do this?
>>> 
>>>  Thanks,
>>>  Irvin
>>> 
>>> 
>>> 
>>> 
>>> 
>> 
>> 
> 
> 

Reply via email to