What's the best low power AMD card with the open source linux drivers?
I'm looking for a card to host in a vga passthrough set up, and as AMD has the best open source drivers I thought I'd use one of theirs.
I've got an old Xeon X5687 in an X58 motherboard and an r9 290
Go with something from 2014 or so, it should work well with the drivers but not be too old to do whatever you need done. Avoid anything brand new. For example, when I bought my card, the r7 260x, I couldn't do anything 3D related on it for over a year and now it works very well.
>>53741844
>best foss drivers
what..
You need something with GCN1.2 for full AMDGPU support
Sure radeon is better than AMDGPU at the moment, but it's being deprecated and replaced by AMDGPU.
I got an R9 380X because I wanted the most performance without going to a Fury.
An R9 285 would be the cheapest you can get though a 380 would have lower power consumption at idle. (less than 5W idle)
If you want something just for the host, get the cheapest thing you can find second hand with low power consumption.
>>53741801
Fury Nano
>>53742089
>Xeon X5687
>Fury
I don't think that's in the right price bracket...
>>53742073
What do you think is the sweetspot in terms of maturity of driver vs age of card? I'm looking for something which is v. stable for the host, not very powerful, but not completely depreciated
>>53742175
Well you're going to need GCN for continued updates, as Terascale is deprecated, so at least HD7730/R5 240.
GCN isn't even 3 years old yet, so there's not really many differences between the cards, although do be aware that the AMDGPU driver (for GCN1.2) is a little glitchy at the moment, but is definitely the way forward into the future.
You can still use the proprietary fglrx driver with the 285/380/380X and it's perfectly stable.
Keep in mind that FOSS drivers aren't the be all and end all as the firmware on the card is proprietary.
I would get a second hand 285 or 380 and run fglrx until AMDGPU is stable.
If you are committed to the FOSS radeon driver, you might want to consider an R7 240-260.
>>53742175
Kaveri APUs, 7xxx/2xx GPUs. You need to blacklist radeon in your initramfs and load it in your module config, and initialize vfio or pcistub with the device you want to pass through in kernel boot line or initramfs.
My machine is weird in that I simply cannot allocate more than 3gb RAM AND passthrough together, so I just made a page file drive on a ramdisk.
I saw no benefit from hugepages
>>53742647
I assumed he was passing through the 390 and using the lower power card as the boot host.
But, to add to this, if you are using a GCN1.2 card, you need to pass the UEFI bios file to the VM in order to be able to reboot the VM without rebooting the whole machine.
>>53742704
so is it easier to reboot the vm with a non UEFI card? i'd read that you have to anticipate that but i wasn't sure if it makes things more difficult
>>53742731
No it's just a bug in the GCN1.2 cards.
All of the GCN cards have UEFI bios, but only the GCN1.2 cards have this issue.
The problem is that GCN1.2 cards dont get shut down properly when the VM halts, so it can't be reinitialized without being rebooted.
If the GPU firmware (UEFI bios) is passed to the VM as a file, it doesn't affect the running state of the physical card, and the VM can be rebooted normally.
>>53742832
https://wiki.debian.org/VGAPassthrough#QEMU_arguments_.28no_GUI.29
This line-device vfio-pci,host=01:00.0,bus=root.1,addr=00.0,multifunction=on,x-vga=on,romfile=$HOME/Asus.HD6850.1024.110616.rom \
shows how the card gets passed with its UEFI bios file.
You can download it from techpowerup, or read i directly from the card.echo 1 > /sys/devices/pci0000:00/0000:00:02.0/rom
cat /sys/devices/pci0000:00/0000:00:02.0/rom > vbios.dump
echo 0 > /sys/devices/pci0000:00/0000:00:02.0/rom
where /sys/devices/pci0000:00/0000:00:02.0/ is your GPU as listed by
lspci
>>53742832
ahh right. thanks for the info. i'm going to scout out a second hand 7750, they seem to be the cheapest at around £30
>>53743002
Not a bad choice, but the Gigabyte model is prone to coil whine.
>>53743055
noted. thanks again
>>53741801
AMD just doesn't try to prevent you from doing a GPU passthrough like Nvidia
but your platform is too old for proper hardware passthrough virtualization most likely
stop being a freetard and just use the proprietary drivers, AMD isn't going to catch up on linux for at least 2 more years, and by then there's either a chance AMD will have brought out actually competitive cards or will be bought by a competent company that can give them the budget they need to make competitive cards.
>>53743341
i don't need any kind of performance, just stability and compatibility. i've heard amd's open source drivers are best for this, especially when attempting gpu passthrough.
>>53743308
my cpu+mobo support vt-d. are there other things they need to support in order for this to work? what should i check?
>>53743369
You don't keep up with Tek Linux, aka the Wendell channel?
Video on it with a forum post, he mentions in the video that it really is best to use a modern platform, probably some better/other virtualization stuff had been added
https://www.youtube.com/watch?v=16dbAUrtMX4
>>53743341
>install a dev package for any graphical library
>oops, nvidia broke again
No. Fuck shitty binary installers, fuck xrandr incompatibility, fuck manually updating my xorg config yet again because I plugged in a monitor, fuck broken display configs cause kernel updated, etc.
Radeon achieves good enough and "just werks" ootb everywhere.
>>53743416
sweeeeet. cheers senpai
>>53743480
AM3+ should be able to do it well, but probably not worth spending money for a minimal performance gain
