I have provided some incorrect info earlier on debian specifics on virtualization with KVM and Proxmox VE, I'm here to correct it. Also it looks like there is a newer version that I haven't used, so YMMV with all this. If you do go the PVE route and a KVM hypervisor, here's the docs I dug up:
Basically, you'll have a debian install and need the device ID's for all the video cards:
lspci -v |grep VGA
04:00.0 VGA compatible controller: ATI Technologies Inc Device 6719
05:00.0 VGA compatible controller: ATI Technologies Inc Device 6719
0a:00.0 VGA compatible controller: ATI Technologies Inc Device 6719
I posted incorrect paths to VM config files, the actual path is:
/etc/qemu-server/VMID.conf
Full documentation on these options:
http://pve.proxmox.com/wiki/Manual:_vm.confAt the bottom of this page, it describes the format for PCI passthrough of Device IDs:
http://www.linux-kvm.org/page/How_to_assign_devices_with_VT-d_in_KVMIMO, the steps above "VT-d device hotplug" I believe to be unnecessary with PVE.
To reiterate, my functional steps to get working PCI passthrough working in the past with this setup has been:
1. enable VT-d in the BIOS. This will let us create and install KVM virtual machines once PVE is installed.
2. install PVE. It's bare-metal and uses all disk space.
3. Log into web interface, create VM however. Stop it.
4. edit my vm .conf file and add the device ID. For you, this might be something like:
device_add pci-assign,host=04:00.0,id=GPU0
device_add pci-assign,host=05:00.0,id=GPU1
device_add pci-assign,host=0a:00.0,id=GPU2
5. start your VM with new VGA devices detected.
When I say VMID, I mean the standard naming convention in PVE of 101 for the first VM, 102 for the second, and so on with the defaults. One of those links says you can do this hot-plug from a kvm guest console, which is available from web interface or cli. I've never tried that myself.
More links:
http://www.linux-kvm.org/page/VGA_device_assignmenthttp://www.linux-kvm.com/content/pci-passthroug-digital-devices-cine-s2http://pve.proxmox.com/wiki/Tape_Driveshttp://pve.proxmox.com/wiki/Manual:_qmThe qm (man qm)command is your friend for kvm VM management from cli on this distro.
Commands like "qm start 101" are instant win. See this:
-hostpci[n] HOSTPCIDEVICE
Map host pci devices. HOSTPCIDEVICE syntax is:
'bus:dev.func' (hexadecimal numbers)
You can us the 'lspci' command to list existing pci devices.
Note: This option allows direct access to host hardware. So it
is no longer possible to migrate such machines - use with
special care.
Experimental: user reported problems with this option.
So another less supported option, which I have most likely used is to simply issue a command(s) like:
qm -hostpci0 05:00.0
qm -hostpci1 04:00.0
qm -hostpci2 0a:00.0
To add my PCI devices.
http://pve.proxmox.com/wiki/Pci_passthroughThis one details "enabling" PCI Passthrough in PVE 2.0 - which I haven't used yet. This step hasn't been required in the past up to PVE 1.9 to my knowledge, but may be required now. Idk here with this.... this seems new to me, never have performed this step to my knowledge.
Enjoy the info dump homie