The way we've been delivering GPU offering with Cloudstack is by using host tags. So each host with a specific GPU has the host tags, example: a16, and the compute offering with the GPU definition also use the hosttag a16.
We've been using this with XenServer Enterprise and so far , no issue for GPU and vGPU support. Nux: vGPU and GPU are more attractive than ever with AI inferencing workload, GPU for AI and desktop, vGPU for desktop mostly. On Tue, Feb 27, 2024 at 7:00 AM Nux <n...@li.nux.ro> wrote: > This sounds foreign to me, afaik GPU support is limited to certain (old) > NVIDIA Grid cards on Xenserver Enterprise. > Modern GPUs are not supported out of the box, although of course many > here do use them by means of custom xml/groovy scripts. > > How you detect them, no idea, let's see how other users do it, if they > care to share. > > On 2024-02-26 18:00, Douglas Oliveira wrote: > > Hello, > > How does the GPU discovery process work on the hypervisor with SC, > > something similar to what Opennebula does? (through lspci) > > I currently have a service offering created via API for an Nvidia A16 > > GPU, > > which does not work because it is informed that there are no hosts > > available to serve the resource. So I'm unsure whether what doesn't > > work is > > the service offering or the non-detection of the GPU on the host. > > > > Regards >