[vfio-users] RX Vega 56 working fine, no need to sleep and wakeup to reinitialize, just need to tweak the fan

[Tomáš M.]

Except Quentin Deldycke's Vega 64 I didn't see any news about Vega on
vfio-users.

I switched from RX 480 to RX Vega 56 (both Sapphire), I realize I didn't
even needed the upgrade for current system, except some UE4 game where its
stressed highly I am mostly CPU limited (i7 3770), but that is another
story.

When setting things up and being confused that GPU is not in PCI 01:00 but
thanks to its bridges its in my case on 03:00, I did some googling I found
this thread:

https://forum.level1techs.com/t/vega-passthrough-help/119054/3

If I understand it correctly, people have issues with starting the vm and
grabbing the card again once it was already "powered off in vm" and to fix
it, do suspend and wake up.

I thought my issues were due to this as well but its much simpler.

When Arch Host boots, the gpu fan goes over 2000 RPM, being quite loud,
when I launch VM it spins down instantly and driver controls it further.
With defaults setting of the AMD OverDrive the minimal RPM for fan is just
10, which basically means off, as a result when I turn off the VM, after
60s the gpu fan goes 100% (lots of noise with reference design). Suspend
fixes that to go back to 2000 RPM, but there is no way I can boot up the
Windows again, it seems to boot but its just get stuck, and no resets help.

After one or two host resets the gpu works again.

Realizing this, I just set minimal rpm in my overdrive profile to 1200,
which effectively means some 1400 real rpm, which is just enough and
silent. After shuting down the VM, fan speed does not change and no more
air bombs and starting the VM again works just fine, as did with previous
GPUs using this script to launch it all( -usbdevice is deprecated by now):

sudo qemu-system-x86_64 \
>   -enable-kvm \
>   -m 16384 \
>   -cpu host,kvm=on \
>   -smp 4,sockets=1,cores=4,threads=1\
>   -vga none \
>   -rtc base=localtime \
>   -device ich9-usb-uhci3,id=uhci \
>   -device usb-ehci,id=ehci \
>   -device nec-usb-xhci,id=xhci \
>   -device vfio-pci,host=03:00.0,x-vga=on,multifunction=on \
>   -device vfio-pci,host=03:00.1 \
>   -device vfio-pci,host=00:14.0 \
>   -drive
> if=virtio,id=drive0,file=/home/pulec/vm/rawdisk/windows10.img,format=raw,cache=none,aio=native
> \
>   -drive
> if=virtio,id=drive1,file=/motherstorage/KVM/data_games.img,format=raw,cache=none,aio=native
> \
>   -drive
> if=virtio,id=drive2,file=/sssd/KVM/ssd_games.raw,format=raw,cache=none,aio=native
> \
>   -usbdevice host:1532:005c \
>   -usbdevice host:1532:0007 \
>   -usbdevice host:1131:1001 \
>   -usbdevice host:04f2:0116 \
>   -usbdevice host:054c:0268 \
>   -usbdevice host:1d50:6022 \
>   -netdev user,id=vmnic -device virtio-net,netdev=vmnic
>

This still uses SeaBios, legacy booting, no UEFI.

Hope this helps to anyone deciding about this beast of GPU.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20171124/2a81743e/attachment.htm>