Which (Tesla, Quadro) GPUs support 'direct passthrough' for operation within VM (VMWare ESX)

Hi (nvidia) folks,

According to this blog entry - GPGPU Computing in a VM - VROOM! Performance Blog - , certain Nvidia GPUs which support so-called ‘direct-passthrough’ are capable to be used for GPU (CUDA) computing within an Virtual machine (VMWare ESX).

The blog mentions that Quadro GPUs (i support at least the newst Fermi-architecture Quadros) support ‘direct-passthrough’. Is that right ? Furthermore, I would like to know if the Fermi architecture Tesla GPUs (C2050, etc…) do support also ‘direct-passthrogh’ ???

According to the blog, no modifications to the application which uses CUDA have to be made, if I understood correctly. Is that right ?

I assume the Tesla must support this pass through, because some similar technology is used in Amazon EC2 to allow virtual machines (I believe EC2 is based on the Xen hypervisor) to access Tesla C2070 cards. Hopefully someone with direct experience with VMWare can confirm…

Any official information from NVIDIA regarding this ‘direct passthrough’ feature for recent Fermi-based (Quadro|Tesla) GPUs ? Would be quite important for me, in order to not propose the wrong GPU for folks who use our GPU algorithms in a e.g. a VMWare ESX environment.

Hi NVidia.

I don’t know if this is the correct forums…
We need to install a NVidia P2000 (Q0V77A) on a HP ML 360 Gen10 with VMWare esxi 6.5 and pass it in passthrough to a Win2016 Virtual Machine already working.
I found fragmentary information on the internet (no precise and complete official VMWare article) explaining how to configure the video card to be viewed and managed directly by V.M. and the part that worried me is that someone says that the V.M. must be configured with the Bios in EFI mode while the V.M. which I have already operative from the customer is definitely installed in normal Bios mode, being also the mode recommended by VMWare during installation.
The problem is that if i change the Bios in EFI mode, the V.M. not boot anymore.
If the Bios in EFI is an essential prerequisite for the correct operation of the video card, this would mean having to reinstall the virtual machine, the terminal server, the ERP software. Can you give me more precise information?
Thanks a lot.