Create own XenServer vGPU-Types

Hi folks,

just a simple question.

Is there a possibilty to create own XenServer vGPU-Types? I just looked a few minutes at the api of xenserver, but i think its a nvidia related question.

We want the K140Q Type, but with 2 GB of GPU Ram instead of 1 GB.

Any hints, thought would be appreciated and nice.


Currently these are defined at the driver level between NVIDIA and Citrix, so no customization is supported. We do add these thoughts to our feature request list for consideration. I am curious about the use case for this profile, at that level of frame buffer you are looking at only 2 users per GPU, and a K2 may make more sense? With the use case, also what Quadro card are you comparing this to?

cause of security reasons iam not aware about the application, so i cannot say something about it. the tester is only reporting, this card/config is bad or good with lil hints. i gave them the q6000 --> good, grid k2 with k260q in xenserver --> good, k140q --> bad, cause of low memory… just for testing i hacked a lil in xenserver and found a way to get what i want. now i have a custom profile (named it k160q), which is actual the k140q, but with 2gb of ram. windows was complaining about the new pci.ids i used, so i put the new pci.ids in the inf of nvidia with config of k140q. its working and the test is in pending state… if someone is interesting how to do that, i can explain…

qp - I need to stress this is an unsupported approach and not appropriate for production environments. This functionality is enabled and posted in the readme solely for temporary diagnostics and troubleshooting, not for ongoing use.

We are still questioning the need for this profile, what is demanding this level of frame buffer? If they need this they really should also be looking to move the increase in CUDA cores, DDR5, and power of the K2. The K1 is more designed for density verses the high performance needs that the K2 typically answers. Have the tester privately message me and I can bring back line engineers in to help sort out their specific GPU needs.

k all right. just wanna give some feedback about the modified k140q profile. the test was good and the gpu load was even lower as the orig. k140q profile, so i think the driver is splitting the rest of ressources dynamically.

i dont know the app, but i think the need for this use case is only a bunch of ram. definitely more than 1 gb per user.

at the end of the day, the big question is: how many users can i get on which grid that cost $$$$… and i think, its better to put 8 users on a grid k1 instead 4 on a grid k2 (perspective on rack space, servers, etc.)

so my approch at @nvidia do something about the static profiles.