i'm doing GPU passthrough wrong

[leesteken]

Try enabling NUMA (numa: 1) because you have multiple sockets. Start will less memory for the VM. Try passing the whole device (hostpci0: 0000:02:00,pcie=1) instead of only the VGA part of the GPU. Ask NVidia what their drivers need to work properly in this setup, because the passthrough itself is working.

 

[nick.kopas]

Memory usage run at 80% and I can't connect to my VM... Promox is definitely capricious.

Try running these commands in the Proxmox shell before starting your VM.

echo 1 > /sys/bus/pci/devices/0000\:02\:00.0/remove
echo 1 > /sys/bus/pci/rescan

 

[Fox_Mulderr]

Try running these commands in the Proxmox shell before starting your VM.

echo 1 > /sys/bus/pci/devices/0000\:02\:00.0/remove
echo 1 > /sys/bus/pci/rescan

I can't get this command to work.

 

[Fox_Mulderr]

Try enabling NUMA (numa: 1) because you have multiple sockets. Start will less memory for the VM. Try passing the whole device (hostpci0: 0000:02:00,pcie=1) instead of only the VGA part of the GPU. Ask NVidia what their drivers need to work properly in this setup, because the passthrough itself is working.

I confirm when the passthrough works, but ask NVIDIA what their drivers need? What? You want I call them? lol Not sure that’s work. Seriously.... I don’t know. My driver is the last one.

 

[leesteken]

I confirm when the passthrough works, but ask NVIDIA what their drivers need? What? You want I call them? lol Not sure that’s work. Seriously.... I don’t know. My driver is the last one.

Only they know their closed source drivers and what it assumes or requires. Maybe they have a forum, mailing list or (paid) support like Proxmox has? Sorry, I have no experience with NVidia (devices) myself. I believe they always supported virtualization with Quadro's, so maybe they have experience with this.
Have you tried passing the whole device (All Functions in the Proxmox web GUI) with hostpci0: 0000:02:00, with or without ,pcie=1? Some drivers assume all functions and some drivers assume the (passed through) device under a PCIe root, which you don't get (and often don't need) unless you enable PCI-Express.

 

[Fox_Mulderr]

So no way to remove error 43? NVIDIA couldn't tell me in passing

 

[nick.kopas]

So no way to remove error 43? NVIDIA couldn't tell me in passing

Can you dump the output of this script?

#!/bin/bash
for d in $(find /sys/kernel/iommu_groups/ -type l | sort -n -k5 -t/); do
    n=${d#*/iommu_groups/*}; n=${n%%/*}
    printf 'IOMMU Group %s ' "$n"
    lspci -nns "${d##*/}"
done;

Also, is there any interesting in /var/log/syslog after trying to start your VM?