[arch-general] Is ATI more... "compatible"?

[Sean Greenslade]

On Sun, May 04, 2014 at 05:53:21PM +0300, Dimitris Zervas wrote:
> On May 4, 2014 4:27 PM, "Temlin Olivér" <temlin at gmail.com> wrote:
> >
> > > It will just hand it to the VM (windows) via VGA passthrough (either Xen
> > or
> > > KVM will be used).
> > > Is that impossible to do with my NVidia?
> > > If it is possible, is it easier with the ATi equivalent?
> >
> > It was already answered.
> > You would be handing the PCI device over to windows (since you have
> another
> > dedicated to linux), so the make does not matter in this case. If you can
> > accomplish that with your current card, than chances are you can do it
> with
> > any other. What matters is the kernel (and probably motherboard) support
> > for the passing of raw PCI devices, so a better question would be whether
> > to use Xen or KVM and on what motherboard.
> >
> > - -Oliver Temlin
> I remember that the kernel had problems disabling the card (99% my fault)
> but generally, after googling my problems back then I saw a kinda
> better/easier way for the ATi
> I also remember that you had to make your nvidia chip look like titan
> (firmware modding), or something like that...

Read through the xen-users mailing list archive. There's been much
discussion on graphics card passthrough and many of
its...idiosyncrasies. The particular thing with NVidia cards is (if I
recall correctly) that the Geforce drivers do not like handling the
memory offsets that are necessary with "virtual" PCIe space. The
Quadro/Tesla drivers are OK with that, however, and it just so happens
that most of the Quadro/Tesla cards share a (binned) chipset with the
Geforce lines. So basically, you take a Geforce and mod it (hardware or
software, depending on the card) to trick the Quadro drivers into using
it.

Definitely do your research and confirm that the card you want to buy is
properly supported with Xen passthrough.

--Sean