XenServer Passthrough GTX 1080

Hello,

I would like to passthrough my GTX 1080 to a VM using XenServer (XenDesktop).
Is this possible with a consumer card?

Regards

BlvckBat

Hi

Yes, this works.

If this is for R&D or a Lab etc, then great. But if you’re planning to use it on a larger scale in a datacenter for enterprise type workloads, then please read the NVIDIA EULA first … :-)

Regards

Ben

Are you sure ?
That is something new in NVidia driver. Every previous attempt failed with "Code 43" in guest driver.

https://stackoverflow.com/questions/47052351/nvidia-gpu-passthrough-with-xenserver-error-43
https://forums.geforce.com/default/topic/1003402/gtx1000-vm-passthrough/

There is some workarounds for KVM (hiding hypervisor) and some for XenServer (https://lists.xenproject.org/archives/html/xen-devel/2016-07/msg01716.html),

My opinion: It is possible to enable GTX 1080 as vGPU card (“turbo” Tesla P4) and use vGPU P4-* profiles for VM (https://gridforums.nvidia.com/default/topic/8934/) because all chip function are the same for the same generation (GMxxx, GPxxx, TUxxx, GAxxx …) (only some parts hw disabled due to chip manufacturing defects or scale down chip). All advanced functions are throttled in driver by “PCI-ID”.

Hi MC

Yes, have done it myself about 2 weeks ago. Physical workstation with XS 7.5 installed and GTX 1080 TI. Requires BIOS modification so that XS doesn’t grab the 1080 for its self, but works absolutely fine :-)

Regards

Ben

Hi BJones, can you share some details how you did it?

Guys,
this is not the right place to discuss topics that violate our EULA. For sure there are ways to do so but this is not permitted as we don’t allow to use Geforce in virtualized environments.

Regards
Simon

.

Simon,
Any license agreement that restricts the personal use of an overpriced piece of hardware that I bought is illegal.
Your EULA only concerns the use of your software, not your hardware. And it says ‘no datacenter deployment’, not ‘no virtualized environment’. My home is not a datacenter.
It means that we have the right to passthrough our gpus on our own virtualized system, and to perform any tricks that will override the artificial limitations of your software, as long as we’re not in a datacenter.
It’s my understanding that NVidia cannot imagine a retail consumer doing something else with their gpus than playing video games, and that’s why this lazy trick is implemented in Nvidia software to block companies from buying GeForce for their datacenters, not concerned about the implications for other legitimate retail consumers use cases, and the potential for an expensive class action down the line.
Additionally, going so far as to block any discussions on this topic on your official forum is a bit insulting for the people that spend so much money buying your products.
Sincerely
Paul