Graphics card options for Linux host (GPU Passthrough)

So with ryzen firmware update that seems to fix IOMMU and looming release of vega upgrading my GPU Passthrough setup seems like a good idea. The only problem at least theoretical problem is that there is no IGPU in any of the Ryzen cpu. With Skylake cpu which I use now, Linux host uses the iGPU and it works fine, it is more than enough horsepower for the host since the most demanding task is video playback.

With that what are my options to providing the Linux host with a gpu? Here are my requirements / preferences:

-4k@60hz support
-DisplayPort and or HDMI output (via adapter might be ok too)
-Low cost
-Low power and noise (as close to iGPU as possible)

Something like RX480 or even RX560 seems like total overkill bot in price and in power usage. Any lower end cards normally dont support 4k resolutions. Also is there a way to roughly compare the additional power consumption of a system with a Dedicated GPU as opposed to 6700k iGPU?

Am I missing any options? Is there perhaps something out of the box like USB graphics cards that could work…? Any suggestion much appreciated!

Technically the Bristol Ridge APUs are also called Ryzen well actually intensifies:smile:

To the topic though, I’m somewhat in the same boat. All the really low cost cards either don’t support 4k, or don’t have more then 2 digital connectors, or both.

I was even thinking about something like the FirePro 2460 if it were cheap, but new it’s almost as costly as a “gaming” card… sooo, there’s that (also that specific card has some caveats regarding displays, seems to be rather picky).

Also found the GPRO 4200 though I can’t find which chip this is exactly or if it’s just an old card.

Though for the host just for a single 4k display even a R7 250 or R5 230 should be enough, no? On the nvidia side 1030 or something… Or do you want multiple displays?

What is your passthrough card? You’ll generally get better compatibility and less setup headaches if you use a card from a different vendor than your host.

Assuming you want a vega card for the VM, you’d probably want to go with a 750 Ti, a 1050, or a 960.

If 4k60 on the host is an absolute must, the cheapest option is a Quadro 410 ($30 or less,) or a GTX 760 if you can find one:

from the nvidia site:

“All GeForce GTX 600 and 700 series GPUs can support 4K resolutions through DisplayPort.”

4k60 started with the 760+ and quadros from the 410+ (750 Ti is newer than the 760, also has support)

if you go with the 410 the additional power draw will be negligible.

another option is to go with a matrox MX card

I could live with single display and single digital output. Display port would be nice as I could just use DP to HDMI adapter for now (current display only has 1 DP connector). The biggest issue with old cards is the power draw. Yes they seem quite cheap but and it seems that even R7 240 supports 4k@60 but they are older architecture and will draw more power. 5W of additional power consumption is at least an extra 5 euros in electricity a year not to mention cooling and fan noise. I’ll have to look into how much extra power it draws compared to iGPU. If it only 5w extra then I can live with it but if it’s 20w it seem to wasteful to me. I will probably have to compare the whole system power draw since the cpu is different too.

I use RX 480 now. I dont really think same vendor cause real issues once you know what you are doing to be honest. Once setup it will work just the same anyways.

Quadro 410 seems interesting. If there are no software issues or display to hdmi adapter issues then it looks like an excellent option. Only other thing is power draw especially since this is an old pro card where power saving features might not have been important.

I dont see any current maxtor cards with video out on their website.

It doesn’t matter as much on platforms with proper IOMMU isolation, but it gives you a lot more flexibility on those that don’t (e.g. ryzen, skylake) For example, some AMD cards will attempt to communicate with each other despite ACS patching, and similar models of nvidia cards will need much more rigid configuration or force you to use seabios. It’s simpler just to avoid all those complications by using 2 different chipset vendors, because it’ll pretty much always work on anything without having to waste time on configuration and breakages.

It’s like a condom. Not necessary strictly for the activity, but it provides a much lower potential for mistakes (and dealing with their consequences.)

If power is such a concern that a 410 draws too much, maybe look at a 1030, but the 410 is a 35 watt card under max load. Last I checked that’d put a ryzen 7 system just over like, an i7 7700k under load.

matrox, not maxtor. They’re business cards that pretty much only exist to drive displays. Their MX9 series is 4k enabled afaik and you can pick them up cheap on ebay

I am not really worried about maximum power draw becasue it is almost never going to get there. It’s the idle / light use power draw that’s a concern. I dont mind if it’s a 300w max power card as long as it has good idle power state. Quadro does seem like a much better option in this case compared to R7 240 or example.

matrox… such a similar name for a similar company :slight_smile:

idle consumption seems to be around 9w, so not as good as new low end cards cards (5w or so) but not bad either.

OK I do understand your point, however do also consider that the iGPU uses some power as well and if you deactivate it (or there is none) that should equal out.

I cant find any info on idle power consumption for quadro 410. If you have any resources please link to it. Assuming 5-year life cycle its worth paying at least 25 euro extra assuming everything else is the same. More recent architecture would be preferred of course for other reasons too such as better video decoding and driver support.

Also when I say idle I dont mean with monitor off but rather just normal desktop use which is the state it will be in mostly.

I would be nice to know the additional power skylake iGPU might consume and how much something like quadro 410 could consume for comparison. It’s hard to compare of course since it’s really ryzen with quadro 410 vs skylake with iGPU comparison.

A perfect card would of course be high efficiency, low idle power draw and have x265 acceleration :slight_smile:

Best I can do is the amperage off the power section off my riser cable:


.7A 12v is about 8.5 W

Yes it seems there’s just no detailed review info for quadro 410 which is not surprising seeing as it is both rpo card and very old by now. I will have to compare my current skylake system with ryzen system I plan on getting. The quadro will be unknown until I buy it I guess.

it works fine for driving displays.

Then again pretty much anything works for driving displays

Well, power consumption will be inherently different for different video cards. I will try to looks for similar consumer level card from that time to compare power usage which should be similar to quadro cards of that time. Quado 400 also seems to support 4k and is even cheaper so I might go for that.

http://www.pcgamer.com/amd-returns-to-the-high-end-graphics-card-market-with-rx-vega/ *

“The Vega also includes improved support for professional level features, things like SRIOV (for hardware virtualization).”

Test it out (when its out), and post tripreport here😀

  • if this article is trustworthy…

Wendell ruled out SRIOV in Vega quite recently I believe. It also just seems too good to be true so its probably not included. will have to keep an eye on it.