[vfio-users] (good) working GPU for passthrough?

[Paige Thompson]

Hey there,

I second Bronek with what he said about the correct pcie configuration, you will have to make some edits in virsh edit as well (virt-manager is limited in what you can do with the pcie bus.) As far as what type of card it really shouldn't matter, in theory but depending on the guest OS and drivers your bus configuration may need to resemble something that he driver will work with. I have a copy pasta that covers a lot of work I did to troubleshoot such a problem, there's also a libvirt xml in this page for reference: https://gitlab.com/snippets/1788426

However if I could suggest something, I really want to find a cheap mxm3 (laptop graphics) to pcie adapter card that is single slot like the one Eurocom makes: https://www.ebay.com/itm/MXM3-to-PCIe-NVidia-Riser-Card-for-MXM-3-0b-3-1b-NVIDIA-GPUs-for-desktop-PCs-/132503008933

 if like me you desire something that doesn't need to use a whole two io slots and blocks off a usable pcie slot I'd look into that or something like it for my next card(s). Unfortunately Eurocom knows they have a niche market and the cost is indicative of it. $400 for a breakout / conversion adapter seems way to steep for anybody let alone me. I'm also not sure whether or not they make IO shields for them, which sucks if not. Everything about the Eurocom cards looks pretty decent, lots of breakouts and there are heatsink / fans. Although the mini dp headers bother me because they don't look like they're being held on by anything except the solder leads and I'd hate to have something like that break off my card.

I really like the idea and I don't mind too much because most of the graphics adapters you can get for laptops now are more than good enough, presumably you can make use of every single x16 slot in your system with cards like that as well.



________________________________________
From: vfio-users-bounces at redhat.com <vfio-users-bounces at redhat.com> on behalf of Georgios Amanakis <gamanakis at gmail.com>
Sent: Friday, January 18, 2019 6:41:30 AM
To: Bronek Kozicki
Cc: Kash Pande; vfio-users at redhat.com
Subject: Re: [vfio-users] (good) working GPU for passthrough?

I am using an Asus DUAL-GTX1060-O6G which doesn't suffer from the reset bug. With the usual workarounds, I have never suffered Error 43 (using either the official NVIDIA driver in Windows, or Asus own driver). I believe the necessity of the ACS patch depends on the motherboard and the underlying IOMMU groups. I am using a workstation X9SRA where every PCI-e slot sits alone in its IOMMU group (admittedly old, based on chipset C602, X79 era) and never had any problems passing through the aforementioned GTX 1060, or the MSI GT 710 2GD3H LP (low-end).



On Fri, Jan 18, 2019, 8:40 AM Bronek Kozicki <brok at spamcop.net<mailto:brok at spamcop.net> wrote:
>
> On Thu, 17 Jan 2019, at 10:39 PM, Kash Pande wrote:
> > On 2019-01-17 2:03 p.m., Tobias Geiger wrote:
> > > I tried a VEGA 64 - i was able to live with the ACS patch needed here
> > > (Z170 Chipset... not needed with the old HD7800, but whatever...) -
> > > but i couldn't stand the reset/FLR/Bug which forces you at least to
> > > suspend/resume the host when you want to reboot only the guest...
> >
> >
> > AIUI, this does not happen using the Q35 machine type with proper PCIe
> > root complex heirarchy.
> >
>
>
> I do not have great experience with "Q35 machine type with proper PCIe root complex hierarchy". Selection of the chipset type Q35 in the virt-manager upon creation does not seem to work with UEFI type of firmware (as recommended for vfio passthroug). Or perhaps I should have downloaded a more recent Tianocore UEFI firmware for my virtual machines?
>
>
> B.
>
>
> --
>   Bronek Kozicki
>   brok at spamcop.net<mailto:brok at spamcop.net>
>
> _______________________________________________
> vfio-users mailing list
> vfio-users at redhat.com<mailto:vfio-users at redhat.com>
> https://www.redhat.com/mailman/listinfo/vfio-users