vgpu_unlock

One example of how they might be abusing their monopoly is by forcing data centers to pay an inflated price for hardware that is similar to consumer GPUs: https://github.com/DualCoder/vgpu_unlock

Wasn't it just a very expensive GPU which basically had a flag that enabled the feature in the driver? I seem to recall reading about people unlocking their NVIDIA GPUs to allow the nested virtualization. --- Edit: here it is https://github.com/DualCoder/vgpu_unlock

This isn’t even the first time a hacked driver has been used to unlock some HW feature - https://github.com/DualCoder/vgpu_unlock

Hardware segmentation in general is just pretty unfortunate. Another example I saw mentioned recently was in-band ECC - Intel’s memory controller apparently supports this but is only enabled in some limited scenarios [0]. If I could enable in-band ECC on my workstation I absolutely would. Given that some folks did manage to unlock the VGPU functionality on consumer GPUs [1] the silicon is certainly capable, but I guess Nvidia doesn’t want to cannibalise their data centre sales. [0] - https://edc.intel.com/content/www/us/en/design/ipla/software... [1] - https://github.com/DualCoder/vgpu_unlock

The vGPU feature can be unlocked for several Nvidia GPUs: https://github.com/DualCoder/vgpu_unlock I did not yet try it myself (maybe it doesn't even work anymore?), and it might not be an option for businesses.

I'm glad Intel has been swallowing their pride on this one, I want them to succeed for once. What really would drive me to support them directly would be (para)virtualization support via SR-IOV/GVT-g. Especially given Microsoft's recent stunts an increasing amount of people (including me and close friends) are evaluating moving to Linux while keeping a Windows VM around, but both NVIDIA and AMD seem adamant about keeping those features gated to enterprise cards (even thought it's a software thing, see community efforts to reverse this[1]) so they can charge ludicrous prices to businesses for those. My opinion is that if Intel wants to get their foot in the door they should win over enthusiast by breaking patterns such as this one rather than try to compete exclusively on performance. [1]: https://github.com/DualCoder/vgpu_unlock

One issue with device virtualization is that the required hardware features are often locked out in non enterprise-grade hardware (although they can sometime be unlocked, for example: https://github.com/DualCoder/vgpu_unlock).

> Is that done at the driver level? No, the FP64 units aren't physically present on silicon in high numbers on the non xx100 dies. However, limitations enforced just by the driver and its FW set: - GPU virtualisation (see: https://github.com/DualCoder/vgpu_unlock) - NVENC video encoder limitations to 2 simultaneous streams on customer cards - Lite Hash Rate limitation enforcement to make GPUs less attractive for miners

Anyone interested in this setup try this if you have NVIDIA? https://github.com/DualCoder/vgpu_unlock Looks like it can enable a virtual GPU so you don't need to actually use 2 GPUs, but I haven't tried it, it looks pretty cool if you only want to run only 1 to switch.

Is the process of unlocking these features on Nvidia GPUs similar to something like the vgpu_unlock tool is doing?[1] No affiliation, just came across it trying to find a replacement to the deprecated RemoteFX vGPU and am out of my depth. [1]https://github.com/DualCoder/vgpu_unlock

I use https://github.com/danielfullmer/nixos-nvidia-vgpu + https://looking-glass.io/ very successfully Can theoretically game on Linux and windows at same time with just a single gpu (2080 ti), mainly use it cause I only have single gpu and I don't want to loose hardware acceleration on Linux, the nixos module both applies a the vgpu unlocker for consumer graphic cards (so u can split a gpu in multiple virtual gpus) https://github.com/DualCoder/vgpu_unlock And merges it with the GeForce drivers so the host gpu does not stop having display output And I just passthrough the vgpu and Xbox controllers to the vm with qemu and my main windows nvme disk (windows naturally just works inside and outside a vm for me, dual boot)

That's not NV related sadly, but applies for all Android TV devices. It's beyond their control... And for PCIe VFs for virtualized GPUs, AMD does _not_ provide it either outside of their datacenter line. Note that if you just passthrough the whole GPU, that works out of the box on NVIDIA cards. (btw, for some NV cards, there is: https://github.com/DualCoder/vgpu_unlock to crack that protection)

It depends on your internal GPU. If it's an Intel GPU, you could use vfio-mdev, which allows you to split your physical GPU into smaller counterparts. It will only work with the macOS version that are compatible with Intel GPUs (probably most of them). Have a look here for a guide on Archlinux : https://wiki.archlinux.org/title/Intel_GVT-g. And here for a script to deploy macOS on KVM : https://github.com/kholia/OSX-KVM. Side-note : vfio-mdev can now be unlocked on consumer-grades Nvidia cards too : https://github.com/DualCoder/vgpu_unlock. Sadly, it is not possible for AMD-cards. By the way, I am working on Phyllome OS (https://phyllo.me/), which is an attempt to make it easier to do such things. But please don't tell anyone :)

Perhaps, but the standard VFIO-style passthrough methods (if you can call anything about that "standard") is still a terrible solution; it requires more than one GPU to use both systems at once, requires navigating horrifically underdocumented driver systems that brick the system when experimented with (just like everything in Linux does, for that matter), and requires at the very least switching inputs between the host and the guest even if all of that works perfectly somehow. Sure, there might not be any more Code 43 (which wasn't the hardest thing to bypass, but you did have to hide the fact that the VM was a VM), but Code 43 was the relative tip of the iceberg compared to all the other setup you had to do to even get there in the first place. Passing through an nVidia GPU the way that data centers do it, where you can run several hard