Has anyone tried PCI passthrough on a Linux host to a Windows guest?
I want to run Windows in a VM and give it ownership of the discrete graphics card.
How are the framerates? Is Windows crashy in a VM? Is Arch the only suitable distro? Because while Arch is nice, I like the stability that Debian offers.
Damn. I was looking into this earlier.
Here's a bump, I wanna know more about this.
>>52025316
I really wanted to try this out but it's not supported by my 3570k. The non-k 3570 supports it though, which I find to be bullshit
I currently do this, and it works great.
The framerate appears to be more or less the same as bare metal.
Pretty much every guide I've seen suggests using rolling-release or bleeding-edge distros. I use Fedora 23 (with virtualisation packages from rawhide) and haven't had any problems yet.
Also, what graphics card are you getting?
>>52025522
I think there's only a handful of K Intel processors that support VT-d.
>>52025316
There's a bug in the newer Linux kernels that breaks this. I don't remember much but watch out for that.
>>52025522
I never liked the K processors, a lot of useful things are disabled. It's only good for normal office or gaming.
>>52025316
>How are the framerates?
Almost exactly the same
>Is Windows crashy in a VM?
Honestly I think it's more stable in a VM
>Is Arch the only suitable distro?
No
>>52025522
>>52025638
I believe the Haswell refresh enables VT-d on all K models
>>52025638
I'm planning to get a chip with integrated graphics and let Linux use that, then give my GTX 760 to Windows.
I'm worried about breaking bleeding-edge distros, because now the host is a single point of failure. If it dies, it takes Windows with it.
I'm quite paranoid about updating such distros.
Reporting in
>>52025708
Use 4.0.9 from the kernel ppa.
>>52025638 (followup)
Felt like sharing my hardware setup
CPU: Intel Xeon E3-1226 v3
GPU: AMD Radeon R9 290X (ASUS DirectCU II)
MB: Supermicro X10SAE
RAM: 2x8GB non-ECC
>>52025786
>chip with integrated graphics [for] Linux
This works great (providing you use EFI to boot the Windows guest)
>GTX 760
Nvidia is a bag of dicks when it comes to PCI passthrough support. You have three options:
1) Use a known-working AMD or Nvidia Quadro card
2) Don't expose hypervisor to guest and disable hyperv enlightenments (reduces performance)
3) If possible, mod your card so that it's detected as a Quadro card.
>>52026864
I don't have a Xeon or a Supermicro board. Will a consumer i7 with an Asus Z87 Deluxe board suffice? Is the Vt-d passthrough a server-only thing?
>>52027285
Which i7 in particular?
Not sure about the motherboard you have. Does it have an option in the BIOS setup to enable VT-d?
>>52027318
I haven't bought the board yet, only have the chip because it was recycled from another dead desktop, an i7-4770T.
I'm still in the research stages of building, but manufacturers don't list stuff like that on their websites. I'm only relying on word of mouth.
I went through a few of the steps on the arch wiki but it seemed to sort of drift off at a certain point and get super ambiguous
running a GTX 660, i5 4690 and a mobo with vt-d support, but from what I've been researching I dont think a 660 is going to work for it.
>>52025522
That's the reason I bought a 3550 when it came out desu senpai.
Newer K CPUs support VT-d.
>>52025316
>Has anyone tried PCI passthrough on a Linux host to a Windows guest?
I'd just like to interject for a moment. What you’re referring to as Linux, is in fact, GNU/Linux, or as I’ve recently taken to calling it, GNU plus Linux. Linux is not an operating system unto itself, but rather another free component of a fully functioning GNU system made useful by the GNU corelibs, shell utilities and vital system components comprising a full OS as defined by POSIX.
Many computer users run a modified version of the GNU system every day, without realizing it. Through a peculiar turn of events, the version of GNU which is widely used today is often called “Linux”, and many of its users are not aware that it is basically the GNU system, developed by the GNU Project. There really is a Linux, and these people are using it, but it is just a part of the system they use.
Linux is the kernel: the program in the system that allocates the machine’s resources to the other programs that you run. The kernel is an essential part of an operating system, but useless by itself; it can only function in the context of a complete operating system. Linux is normally used in combination with the GNU operating system: the whole system is basically GNU with Linux added, or GNU/Linux. All the so-called “Linux” distributions are really distributions of GNU/Linux.
>>52025316
>How are the framerates?
optimally 95-99.9% of original depending on individual setup and hardware
>Is Windows crashy in a VM?
depends on individual setup and hardware
>Is Arch the only suitable distro?
no, in fact fedora is generally the easiest distro to do it on because the recommended repos and libraries are fedora ones anyway
the issues with pci pass-through generally stem from motherboard manufacturers not supporting it properly or at all (namely asus, ymmv)
there's some minor issues such as nvidia cards requiring quadro firmware to be flashed for some nvidia cards to work(nvidia gimping consumer cards), and sometimes having to unmount graphics cards from the guest os (windows 8.1 at least shows this in the hdd unmount tray icon automatically) before shutting the guest os down before the card can be used in a vm again
oh, and as mentioned >>52025708
>There's a bug in the newer Linux kernels that breaks this. I don't remember much but watch out for that.
don't use newer kernels until it's fixed
>>52025708
>I never liked the K processors, a lot of useful things are disabled. It's only good for normal office or gaming.
high end k processors like devils canyon or haswell-e have vt-d, it's only the consumer grade (at least, up until skylake) where vt-d was disabled on k chips
welcome to chip binning and disabling features to sell more silicone
>>52025786
use fedora, it's not going to break (vt-d might, with newer kernels, but that's a kernel bug that affects every distro)
>>52027398
>but manufacturers don't list stuff like that on their websites
yes they do
http://ark.intel.com/products/75125/Intel-Core-i7-4770T-Processor-8M-Cache-up-to-3_70-GHz
ctrl+f vt-d
4770t is lga 1150, looking up chipset support for vt-d
https://en.wikipedia.org/wiki/LGA_1150
ctrl+f vt-d
you'll need q87 for vt-d
from intel's site:
q87: http://ark.intel.com/products/75007
z87: http://ark.intel.com/products/75013
ctrl+f vt-d
>>52025316
>Has anyone tried PCI passthrough on a Linux host to a Windows guest?
Yes, this is my daily setup on my home PC.
>How are the framerates?
Indistinguishable from native.
>Is Windows crashy in a VM?
No, but I've heard Nvidia drivers can be.
>Is Arch the only suitable distro? Because while Arch is nice, I like the stability that Debian offers.
I tried to do this on Debian but couldn't get the ACS override patch to work (necessary for spoofing devices isolation for devices in the same IOMMU group for non-Xeon CPUs). Tried Arch instead (which I use on my laptop anyway) and it was much easier.
I recommend a 'bleeding edge' distro. Arch is good, but some of the virtual UEFI dev for QEMU (OVMF) was originally intended for Fedora Rawhide, so you could try that.
Here's my setup:
Host OS: Arch w/ patched kernel 4.2
Guest OS: Win7 64 bit
Host GPU: GTX 760
Guest GPU: R9 Fury
CPU: i7 4790k.
Other things:
My mobo has two USB controllers so I passed one of them through to windows, which I reckon has marginally less latency than QEMU's usb device passthrough feature. Also allows me to switch between OSs quickly with a USB switch.
