Is there anyway someone can hack the drivers or something to “flip the switch”?
1 Like
There have been occasional screenshots I have seen, but nothing that is definitely real, and nobody has said how they did it.
Example:
https://gridforums.nvidia.com/default/topic/8934/nvidia-virtual-gpu-technology/vgpu-on-gtx-or-rtx-or-quadro
https://archive.is/Tq9cg
Guys, have now opened a ticket this matter on NVIDIA forums. It is a place where you can register your interest in this feature for these new geforce / rtx 3080 cards. And communicate that to help to raise awareness back to NVIDIA themselves. Good luck!
1 Like
They know. They will tell you to buy a Quadro.
2 Likes
DANG… I know @wendell did that video series on the AMD Firepros, but I havent found a follow up or any videos on getting Nvidia cards working. In advance I purchased a Nvidia GRID K2. I know my motherboard has SR-IOV support… but I was shy about trying to install it yet. Just getting my home lab how I like it for my first Home Lab Build. I know licencing was and issue, but I was under the impression the olders cards like my K2 didnt need that, till I heard Wendell say they were implementing it retroactively to older cards too. Hell I’ll donate it to him if he want to do a piece with it and tell me how to set it up…lol
if you need help with the card posting due to 4G decoding or any further help please make a thread about it
2 Likes
@GigaBusterEXE
Thank you, I saw his videos but it dednt see he really walked through the steps but rather told us of his proplems without examples if I remember right. I’ll check it out and see for sure. If I do I will post elsewhere for the help. Thank you 
Sorry didnt mean to seem I was hijacking here. Not my intention, just made me think of this project.
The big problem is some motherboards freak out when the PCI ROM BAR space exceeds a certain amount. This is gonna be a issue with consumer boards and SR-IOV:
1 Like
I believe yes, since you can hack/patch the driver to allow NvFBC capture (which is a Quadro feature) on consumer cards. so It is a good sign that “the hardware supports it”.
I don’t understand, why wouldn’t they allow it officially and limit the number of VMs you can run.
You’d think that people who buy or rent a lot of GPUs have more motivation to hack the drivers, so this is effectively only stopping average consumers like us.
2 Likes
This where I get stuck, as I now build a second server to mess around with the older Nvidia K2. The ASRock Rack x470d4u has both SR-IOV and immou is effective at separating my cards properly. Just gonna read up see if I can get it to work. Would be nice to play with the different distros with a little gpu love lol.
Look foward to seeing how this project works out. I plan to start a build thread thank you @GigaBusterEXE. Havent decided to bite the bullet on epic or threadripper yet. Or a full server blade for that matter… Wife would be very upset even if it was just in my office on the FAR side of the house, and AZ is to hot to put a cabinet in the garage lol
@wis.am I agree, but it is a business model. Doesn’t have the same open feeling of support as I am now feeling with the open source community since making the switch away from windows.
Sorry, I am a bit late to this thread, but I have a question. I plan to use a 3090 on linux and run my applications in a docker container. I have previously been able to pass a Quadro GPU device to docker while starting up a container. Does this require SR-IOV support? Will this work on a 3090?
You’re not really virtualizing the GPU that way, so no I don’t see how it would require SR-IOV or passthrough.
It’d be nice for me at least I’m really getting tired of fucking around with VFIO garbage. I’m about to just install windows again and forget the whole thing.
We need to crack the firmware blobs open [unpack them to a human readable state] then determine how to defeat a hash self-check with a hash collision (presumably the card knows what the EPROM memory hashes to). Intel ME’s blob “integrity protection” was defeated with some kind of hash collision using a tool called me_cleaner. If a similar approach could be applied to the blobs read out from the consumer Nvidia or AMD cards then it might be possible to make changes to the blobs that go undetected. That’s just a working theory. There might also be code signing involved so if that’s the case there may be an additional layer to defeat. I could just be barking up the wrong tree entirely but I think I might be onto something so I’m going to chase this lead as far as I can go until I hit some really hard dead end. Seems like the better option compared to letting vendors basically disable features arbitrarily with zero pushback.
1 Like