Is it possible to use original Nvidia drivers when running windows through KVM?

Question

NVIDIA

I'm running Windows via KVM. I install the original Nvidia drivers in the virtual machine (for Windows). Will the card work properly?

Answer 1

It will, but on the way to this there are many "Ifs", basically everything rests on iron. Simultaneous support is required by both the motherboard and the stone (for the Intel platform) VT-d, VT-x + Above 4G Decoding + a second video card (integrated will do) + 1-2 monitors (depending on the number of video output ports, in order to give a weak card under host, and a gaming one for a virtual machine). For AMD, there are a few options. At the same time, there are no guarantees that it will turn out without problems and the first time. Look for similar configs on the network that worked. A small nuance will help to determine the depth of the problems, which can be managed programmatically, but I decided that it would be easier to change the mother, because. the previous one from Gigabyte did not meet this requirement:

1) Run the "dmesg | grep ecap" command.
2) On the IOMMU lines, the hexadecimal value after "ecap" indicates whether interrupt remapping is supported. If the last character of this value is an 8, 9, a, b, c, d, e, or an f, interrupt remapping is supported. For example, "ecap 1000" indicates there is no interrupt remapping support. "ecap 10207f" indicates interrupt remapping support, as the last character is an "f".
Interrupt remapping will only be enabled if every IOMMU supports it.

In principle, mothers from ASRock are praised for forwarding, they, as a rule, have the necessary parameters in BIOS, in general, BEFORE buying, you need to look at what this motherboard can do and what items are in the BIOS menu. Then, in principle, there are guides. I managed to throw the GTX1070 without problems, although it happens that many people are plagued by error 43, this NVIDIA driver prevents the video card from working correctly in the virtual machine, so that they would buy a prof. solutions for many hundreds of thousands of rubles. At first glance, the rules work, there are no lags and freezes, but there are some nuances, for example, the fact that it is easier to turn off the virtual machine inadvertently ... for some type of equipment, this can be harmful. This must be taken into account. And again, the host is Linux in one form or another (although I met the opinion that it is easier to do such forwarding on BSD)
In summary, if you are ready to waste your time and fool around with upgrading / replacing hardware on a PC, then yes! Basically it works and it can be achieved. If you want something that works out of the box, works on the principle of turning it on and forgetting it, then you should spend your time on something else.

Answer 2

Yes, it is possible, but it needs to be supported by the processor, motherboard and graphics card. This is achieved by forwarding a pci-e device with a video card (there may be several, for example video and hdmi audio). It is also HIGHLY recommended to have an additional video card to run the host system (so that the forwarded video card will not be affected by this host system when the system starts up).
And even in this case, a bunch of rakes are possible, for example, drivers may not be installed by a regular installer (but if the system was started natively and the drivers are installed, then later they will work in the virtual machine).
The reason for the problems is that such entertainment by video card manufacturers is designed for expensive server hardware, which means that artificial problems can be created with cheap hardware.
ps on a windows host system, I'm not sure if it will work out for you, and is it possible in principle, I successfully configured ubuntu for linux using kvm on an old amd phenome 61xx, running windows 7 and toys in the guest system (it worked noticeably slower, 30 percent, but I guess because of the processor)

Answer 3

I install the original Nvidia drivers in the virtual machine (for Windows). Will the card work properly?

No.
This is being prevented. Specially. Because the cost of a card that is "officially" cut through vmware, for example (NVidia GRID K1 / K2), in my opinion, is from 250 thousand rubles.
Here the task falls into exactly two parts:
1. Forward the card to KVM (I mean, of course, qemu + libvirt). Here, a bunch of rake with linux settings, with IOMMU and its groups, with rearrangement of hardware in other slots (it helps for redistribution among IOMMU groups), is possible, but in the end this stage can usually be overcome.
2. Run firewood so that they find the thrown card. Nvidia opposesthe work of firewood in KVM, forcing firewood to give an error and not see the card. There are hacks for qemu, but they don't help. You can try an old map that does not yet have this check in the firewood. They say (I didn't check it myself) that this can happen with ATI.