Best works-out-of-the-box dual-head gfx card?

[Stumpy]

Hi,

I have seen a few gfx card requests for "qubes compatible" and lots of posts about gfx card issues so I wanted to be sure about my purcahse so I will ask a similar question. I am in the market for a graphics card, or more specifically increasing the number of ports as I am going from 2 -> 4 monitors. I figure I am going to need to get a graphics card for this that has a minimum of 2 ports.

I have read quite a few posts regarding problems with NVIDA and AMD/ATI cards/chipsets, but they seem to be my only option. From the posts I saw it kinda seems that the AMD/ATI works a bit better, thoughts?

I am hoping those out there who have had “plug in and it just works” experiences with graphics cards can chime in with their recommendations.

I am also hoping to get something as “future proof” as possible, so it will admittedly be a balance, that said I don’t need bleeding edge either as I am not a gamer and don’t do too much with video (does video encoding count?).

I must have a minimum of two ports Other preferences are:

• I really really really want something that will work outa the box, not really interested in recompiling the kernel as I have some seriously defunt kernel karma
• DisplayPort would be nice but DVI will be fine
• low profile (hopefully not too wishful of thinking for a 2+ port card), my box is actually pretty small.
• Lots memory I guess?
• Passively cooled would be nice.

In terms of expansion slots my box has 1 PCI-E X16 Gen2.0 (supports 3.0? not wholly sure about this) and 1PCI-E X4
Oh, and on this machine I am using 3.2 but will of course upgrade to 4.0 once its final.

Thanks so much in advance!

[Yuraeitha]

On Thursday, March 8, 2018 at 11:30:21 PM UTC+1, Stumpy wrote:
> Hi,
>
> I have seen a few gfx card requests for "qubes compatible" and lots of posts about gfx card issues so I wanted to be sure about my purcahse so I will ask a similar question. I am in the market for a graphics card, or more specifically increasing the number of ports as I am going from 2 -> 4 monitors. I figure I am going to need to get a graphics card for this that has a minimum of 2 ports.
>
> I have read quite a few posts regarding problems with NVIDA and AMD/ATI cards/chipsets, but they seem to be my only option. From the posts I saw it kinda seems that the AMD/ATI works a bit better, thoughts?
>

I might ave missed those posts, but the only way I know of right now that allow discrete graphic cards in Qubes 3.2 and Qubes 4.0 is by passing it directly into dom0, which is considered a big "no no" in Qubes security. However Qubes 4.1. might introduce a way to allow a single AppVM to handle a discreate graphic card without introducing new security issues. But lets wait and see how it goes, we barely got Qubes 4.0 out of the door yet right.

There is also the possibility that eGPU can work in Qubes 3.2. and Qubes 4.0 by using high-speed Thunderbolt connections. But for one, this is expensive, like really, they ask too much for these otherwise simple external PCI-e card readers running Thunderbolt. Also I have never actually seen anyone succeed or even try this, it's highly speculative. Considering if you can pass-through a Thunderbolt port to an AppVM, but not an pci-e port, then in a nutshell, if drivers etc. work properly for Thunderbolt/eGPU, you should have high end graphics in Qubes AppVM's.

BUT, this is really too expensive. I will not by any means recommend you do this. Neither would I recommend you hack dom0 and put your graphic cards directly into dom0. You might be better off just waiting for Qubes 4.1. for this capability.

> I am hoping those out there who have had “plug in and it just works” experiences with graphics cards can chime in with their recommendations.
>

You really don't need discrete powerful graphic cards for Qubes. Some modern integrated ones can even do some pretty amazing graphic feats, and many older ones are pretty good too. You can easily run 4k 48" on some couple of years old mobile intel CPU's. But of course, some are horrible too. Be sure to check reviews and benchmarks. But you're already running Qubes 3.2., so you should have a decent idea about this.

> I am also hoping to get something as “future proof” as possible, so it will admittedly be a balance, that said I don’t need bleeding edge either as I am not a gamer and don’t do too much with video (does video encoding count?).
>

I hear most Linux gavers use GTX 1060 (You might want to double check that, google it etc.). HOWEVER, Qubes can't use these cards right now, and it makes zero sense to add it to dom0 as VM's use graphic driver translations anyhow, which means they can't make full use of the graphic card. Onboard graphics is likely just as good on modern hardware. That will likely be different in Qubes 4.1. for a single AppVM pr. graphic card, but we're still on Qubes 4.0. now.

> I must have a minimum of two ports Other preferences are:
>
> I really really really want something that will work outa the box, not really interested in recompiling the kernel as I have some seriously defunt kernel karma

Try check https://www.qubes-os.org/hcl/ for people who previously reviewed Qubes hardware. Also Intel graphics tend to work the most "out-of-the-box" on Qubes. Modern integrated graphics isn't too bad, although keep in mind new AMD graphics are in newer intel chips now. While they are nice and powerful, support might not be so great "yet", considering it's all new on the market. So you might want to stay clear of too new intel chips, in general also anything too new or too old hardware.

> DisplayPort would be nice but DVI will be fine

I assume it's a desktop we're looking at here. Should be pretty easy to find a motherboard having DVI at least, some probably have Displayport too. Try find some in the list, and then check if they got the Displayport/DVI.

> low profile (hopefully not too wishful of thinking for a 2+ port card), my box is actually pretty small.
> Lots memory I guess?

You can get by with 8GB, though 16GB feels much more relaxing.
You could also get ECC ram, though, it's a bit controversial, some find it important for data integrity, others don't think it's common enough to be considered an issue. Generally, it's a messy discussion. But you might want to take your stance there if you should consider a new desktop machine (requires motherboard/CPU/UEFI also support ECC memory.

> Passively cooled would be nice.

Well should be doable, but I guess you're referring to the graphic cards here. Tbh, don't buy graphic cards now, wait and see if it works in Qubes 4.1., and you'll get better value for your money by not buying it so long in advance.

>
>
> In terms of expansion slots my box has 1 PCI-E X16 Gen2.0 (supports 3.0? not wholly sure about this) and 1PCI-E X4
> Oh, and on this machine I am using 3.2 but will of course upgrade to 4.0 once its final.

Do you meet the hardware recommendations (partly requirements, although it can be bypassed, hypothetically making 3.2. = 4.0. in hardare compatibility, however, only if you modify Qubes 4.0. and I'm really not sure if you need to do more than change all VM's to PV or even HVM mode).

>
> Thanks so much in advance!

oh well, I don't think this is what you wanted to hear.
But do feel free to post those posts you mentioned talking about Qubes pass-through GPU. If anyone managed to do it in Qubes in a secure way, that would be quite interesting. But I suspect it might be dodgy insecure hacks?

[Yuraeitha]

On Thursday, March 8, 2018 at 11:30:21 PM UTC+1, Stumpy wrote:

Also keep in mind Thunderbolt support in the kernel isn't too great yet, and also not too much hardware beyond Apple products support Thunderbolt, although they are making a bit bigger appearance on PC's in 2018, but it's still slow. Generally, Thunderbolt is still a big headache, with multiple of ways that it can fail (although not saying it will fail for certain, just that there is plenty of risk).

[Tai...@gmx.com]

AMD stuff is the best choice, they don't artificially hobble
virtualization on their regular cards and they work out of the box.

The reason things break and you need to re-compile every time you update
your kernel with a new nvidia card in your system is because they make
an effort to slow down the nouveau project and in the case of their
brand new cards entirely block it via hardware code signing enforcement
and not even providing any firmware blobs.

There are a variety of single slot half height low power AMD cards but
if you want something newer the only choice is the more expensive
professional series with a fan - the WX4100 for instance is half height
but not passively cooled.

You can obtain a 54xx passively cooled card for around $30-50 but that
is a quite old chipset (although you can install openradeonbios on it)

[Tai...@gmx.com]

On 03/08/2018 06:24 PM, Yuraeitha wrote:

> I might ave missed those posts, but the only way I know of right now
> that allow discrete graphic cards in Qubes 3.2 and Qubes 4.0 is by
> passing it directly into dom0, which is considered a big "no no" in
> Qubes security. However Qubes 4.1. might introduce a way to allow a
> single AppVM to handle a discreate graphic card without introducing
> new security issues. But lets wait and see how it goes, we barely got
> Qubes 4.0 out of the door yet right.
> There is also the possibility that eGPU can work in Qubes 3.2. and Qubes 4.0 by using high-speed Thunderbolt connections. But for one, this is expensive, like really, they ask too much for these otherwise simple external PCI-e card readers running Thunderbolt. Also I have never actually seen anyone succeed or even try this, it's highly speculative. Considering if you can pass-through a Thunderbolt port to an AppVM, but not an pci-e port, then in a nutshell, if drivers etc. work properly for Thunderbolt/eGPU, you should have high end graphics in Qubes AppVM's.
>
> BUT, this is really too expensive. I will not by any means recommend you do this. Neither would I recommend you hack dom0 and put your graphic cards directly into dom0. You might be better off just waiting for Qubes 4.1. for this capability.

I have used an eGPU via expresscard with an X230 laptop for gaming, you
could also buy an T420/430, W520 (32GB RAM), etc.

It works well and can be attached to a VM if the device supports
IOMMU-GFX which these do although I would suggest a workstation for
x86_64 gaming (in or out of a VM) such as the libre firmware owner
controlled KCMA-D8 ($315) a dual socket mobo supporting crossfire and
the 4386 CPU (equiv FX-8310) that can play modern games at high settings.

If you can't find that the KGPE-D16 ($415) is also a dual socket option
which supports up to 32 cores and 192GB RAM, the socket G34 6386SE is
the best and last owner controlled x86_64 CPU.

[Yuraeitha]

that is quite interesting, so the bandwidth is big enough to handle that over expresscard, no big amount of bandwidth is lost? How did you solve the drivers here? Were there any complexities by any chance, or did it work like normal?

[Tai...@gmx.com]

On 03/08/2018 07:23 PM, Yuraeitha wrote:

> that is quite interesting, so the bandwidth is big enough to handle that over expresscard, no big amount of bandwidth is lost? How did you solve the drivers here? Were there any complexities by any chance, or did it work like normal?

It is the same as attaching it to a regular PCI-e slot.

If your laptop has at least PCI-e 2.0 ExpressCard there is a significant
but bearable performance loss as long as the card has enough video RAM
so that you are not swapping out game textures to system RAM which
requires a lot of PCI-e bandwidth.

You can also use a special program to add one of the mini-pcie wifi
slots to the total bandwidth vastly improving performance although this
would require a laptop case modification for the connector (or removing
the bottom panel of the laptop)

[Stumpy]

On 08.03.2018 23:30, Stumpy wrote:
> Hi,
>
> I have seen a few gfx card requests for "qubes compatible" and lots of
> posts about gfx card issues so I wanted to be sure about my purcahse
> so I will ask a similar question. I am in the market for a graphics
> card, or more specifically increasing the number of ports as I am
> going from 2 -> 4 monitors. I figure I am going to need to get a
> graphics card for this that has a minimum of 2 ports.
>
> I have read quite a few posts regarding problems with NVIDA and
> AMD/ATI cards/chipsets, but they seem to be my only option. From the
> posts I saw it kinda seems that the AMD/ATI works a bit better,
> thoughts?
>
> I am hoping those out there who have had "plug in and it just works"
> experiences with graphics cards can chime in with their
> recommendations.
>
> I am also hoping to get something as "future proof" as possible, so it
> will admittedly be a balance, that said I don't need bleeding edge
> either as I am not a gamer and don't do too much with video (does
> video encoding count?).
>
> I must have a minimum of two ports Other preferences are:
>

> * I really really really want something that will work outa the box,

> not really interested in recompiling the kernel as I have some
> seriously defunt kernel karma

> * DisplayPort would be nice but DVI will be fine
> * low profile (hopefully not too wishful of thinking for a 2+ port

> card), my box is actually pretty small.

> * Lots memory I guess?
> * Passively cooled would be nice.

>
> In terms of expansion slots my box has 1 PCI-E X16 Gen2.0 (supports
> 3.0? not wholly sure about this) and 1PCI-E X4
> Oh, and on this machine I am using 3.2 but will of course upgrade to
> 4.0 once its final.
>
> Thanks so much in advance!
>

> --
> You received this message because you are subscribed to the Google
> Groups "qubes-users" group.
> To unsubscribe from this group and stop receiving emails from it,
> send an email to qubes-users...@googlegroups.com.
> To post to this group, send email to qubes...@googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/qubes-users/41c4eb008e246d2e1850afbc6ec8cf73%40posteo.net
> [1].
> For more options, visit https://groups.google.com/d/optout [2].
>
>
> Links:
> ------
> [1]
> https://groups.google.com/d/msgid/qubes-users/41c4eb008e246d2e1850afbc6ec8cf73%40posteo.net?utm_medium=email&utm_source=footer
> [2] https://groups.google.com/d/optout

Thanks to all for the responses.

I think I was unclear about some parts though.
I don't need the newest greatest, new*er* is good just so I don't have
to get something else in the future but most important is being able to
add 2 more video ports to the 2 that my mobo has
Also, my mobo has 2 already and I am hoping to not have to go the route
of a new mobo quite yet (and are there really that many mobos with 4
ports?)
Also, it was advised not to have it go through dom0 (or something like
that) and I would 200% agree, compromising the security built in seems
to defeat the point of having qubes (I realize others may not entirely
agree), but no, I am hoping for an option where I do not compromise
security, am able to have 4 ports total (2 [or more] plus the two that I
already have), and not have to do much fancy recompiling etc.
I really appreciate the responses and if what I added for clarification
helps anyone with other ideas please please do let me know! Having a 4
monitor setup is sort of a dream of mine ;)

[Stumpy]

Ok, 4.0 Qubes is out so I am more motivated than ever before. I took a
look at the WX4100 which looks fantastic, but perhaps is more than I
need (and more than I can afford).

I can do without the passively cooled part, esp if it will bring the
price down a bit (I am hoping for something in the low 100s USD).

In terms of requirements, I suppose (to simplify things) the main
criteria I have are low 100s USD, low profile, and at least two ports,
though 4 would like the WX4100 would of course be awesome. I am totally
open to used, and am not looking for bleeding edge.

Thoughts? Please!! :)

[799]

Stumpy <stu...@posteo.co> schrieb am Do., 29. März 2018, 03:01:

On 09.03.2018 00:57, Tai...@gmx.com wrote:

> [...]

> The reason things break and you need to re-compile every time you
> update your kernel with a new nvidia card in your system is because
> they make an effort to slow down the nouveau project and in the case
> of their brand new cards entirely block it via hardware code signing
> enforcement and not even providing any firmware blobs.

[...]

Honestly, I think this should be mentioned somewhere within the Qubes Docs, maybe a dedicated section "Not recommended Hardware" or something like "Hardware tries to keep you unsecure".

The only way to stop those brands is to speak about it.

We are also having issues with Nvidia in our bigger desktop virtualization projects as customers need to buy not only the server based graphic card but also licensing and subscription...

"Nvidia, you're fired!"

[799]

[Stumpy]

On 08.03.2018 23:30, Stumpy wrote:

> Hi,
>
> I have seen a few gfx card requests for "qubes compatible" and lots of
> posts about gfx card issues so I wanted to be sure about my purcahse
> so I will ask a similar question. I am in the market for a graphics
> card, or more specifically increasing the number of ports as I am
> going from 2 -> 4 monitors. I figure I am going to need to get a
> graphics card for this that has a minimum of 2 ports.
>
> I have read quite a few posts regarding problems with NVIDA and
> AMD/ATI cards/chipsets, but they seem to be my only option. From the
> posts I saw it kinda seems that the AMD/ATI works a bit better,
> thoughts?
>
> I am hoping those out there who have had "plug in and it just works"
> experiences with graphics cards can chime in with their
> recommendations.
>
> I am also hoping to get something as "future proof" as possible, so it
> will admittedly be a balance, that said I don't need bleeding edge
> either as I am not a gamer and don't do too much with video (does
> video encoding count?).
>
> I must have a minimum of two ports Other preferences are:
>

> * I really really really want something that will work outa the box,

> not really interested in recompiling the kernel as I have some
> seriously defunt kernel karma

> * DisplayPort would be nice but DVI will be fine
> * low profile (hopefully not too wishful of thinking for a 2+ port

> card), my box is actually pretty small.

> * Lots memory I guess?

> * Passively cooled would be nice.

>
> In terms of expansion slots my box has 1 PCI-E X16 Gen2.0 (supports
> 3.0? not wholly sure about this) and 1PCI-E X4
> Oh, and on this machine I am using 3.2 but will of course upgrade to
> 4.0 once its final.
>
> Thanks so much in advance!
>

> --
> You received this message because you are subscribed to the Google
> Groups "qubes-users" group.
> To unsubscribe from this group and stop receiving emails from it,
> send an email to qubes-users...@googlegroups.com.
> To post to this group, send email to qubes...@googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/qubes-users/41c4eb008e246d2e1850afbc6ec8cf73%40posteo.net
> [1].
> For more options, visit https://groups.google.com/d/optout [2].
>
>
> Links:
> ------
> [1]
> https://groups.google.com/d/msgid/qubes-users/41c4eb008e246d2e1850afbc6ec8cf73%40posteo.net?utm_medium=email&utm_source=footer
> [2] https://groups.google.com/d/optout

Ok, how about this. Taiidan suggested the WX4100 (thx!) which looks
awesome but it a bit outside my price range. I found the AMD Radeon Pro
WX 2100
(https://www.newegg.com/Product/Product.aspx?Item=9SIA6ZP5X42735). That
seems to be pretty much the same, or more like a step down but similar
otherwise. Should that work ok with Qubes 4.0? It has 3 ports, is low
profile, supports 4k, etc which I think will suffice for me.
Any confirmation that this should work with Qubes 4.x w/o modifications
would really be appreciated!

[Tai...@gmx.com]

On 03/28/2018 09:01 PM, Stumpy wrote:

On 09.03.2018 00:57, Tai...@gmx.com wrote:

AMD stuff is the best choice, they don't artificially hobble
virtualization on their regular cards and they work out of the box.

The reason things break and you need to re-compile every time you
update your kernel with a new nvidia card in your system is because
they make an effort to slow down the nouveau project and in the case
of their brand new cards entirely block it via hardware code signing
enforcement and not even providing any firmware blobs.

There are a variety of single slot half height low power AMD cards but
if you want something newer the only choice is the more expensive
professional series with a fan - the WX4100 for instance is half
height but not passively cooled.

You can obtain a 54xx passively cooled card for around $30-50 but that
is a quite old chipset (although you can install openradeonbios on it)

Ok, 4.0 Qubes is out so I am more motivated than ever before. I took a look at the WX4100 which looks fantastic, but perhaps is more than I need (and more than I can afford).

It is a sexy card right? I too saw the pretty blue color and it was love at first sight xD even though I know that is an overpriced card.
I don't suggest buying an overpriced card like the 4100 just because it is single slot/half height or w/e I would instead get a 5450 unless you absolutely need something better.

I can do without the passively cooled part, esp if it will bring the price down a bit (I am hoping for something in the low 100s USD).

Yeah it will....you aren't intending for gaming right? I would get a 5450 or 5470 they come in passively cooled and half height.

In terms of requirements, I suppose (to simplify things) the main criteria I have are low 100s USD, low profile, and at least two ports, though 4 would like the WX4100 would of course be awesome. I am totally open to used, and am not looking for bleeding edge.

Get a 5450 for $30 bucks I sugg

[Stumpy]

*gorgeous*, like gfx card pr0n :)

> I too saw the pretty blue color and it was
> love at first sight xD even though I know that is an overpriced card.
> I don't suggest buying an overpriced card like the 4100 just because
> it is single slot/half height or w/e I would instead get a 5450 unless
> you absolutely need something better.

I can certainly do with less, like I said I am not looking for bleeding
edge.

>

>> I can do without the passively cooled part, esp if it will bring the
>> price down a bit (I am hoping for something in the low 100s USD).
> Yeah it will....you aren't intending for gaming right? I would get a
> 5450 or 5470 they come in passively cooled and half height.

Nope, no gaming, just multiple monitors photo editing, occasional video
encoding etc

>
>> In terms of requirements, I suppose (to simplify things) the main
>> criteria I have are low 100s USD, low profile, and at least two
>> ports, though 4 would like the WX4100 would of course be awesome. I
>> am totally open to used, and am not looking for bleeding edge.
> Get a 5450 for $30 bucks I sugg

Thanks (very much) for that. That is def a sweet spot in terms of price.
Because I can't leave good enough alone 8O I started looking around for
something similar, that was AMD, and had the ports I was hoping for
(instead of buying cables like hdmi 2 DVI etc).
I found the AMD-Radeon-HD-6570
(https://www.ebay.com/itm/HP-AMD-Radeon-HD-6570-1GB-DDR3-PCIe-x16-DP-DVI-Video-Card-637184-001-637997-001/232686259800?hash=item362d2e8e58:g:JfIAAOSwY~1Z86qr)
which conveniently has DP and DVI (not HDMI or VGA). Thoughts? (thanks
for your patience/help!)

>
> --
> You received this message because you are subscribed to the Google
> Groups "qubes-users" group.
> To unsubscribe from this group and stop receiving emails from it,
> send an email to qubes-users...@googlegroups.com.
> To post to this group, send email to qubes...@googlegroups.com.
> To view this discussion on the web visit

> https://groups.google.com/d/msgid/qubes-users/fd005b4d-48a4-0c26-dacf-152f61bcb2ed%40gmx.com

> [1].
> For more options, visit https://groups.google.com/d/optout [2].
>
>
> Links:
> ------
> [1]

> https://groups.google.com/d/msgid/qubes-users/fd005b4d-48a4-0c26-dacf-152f61bcb2ed%40gmx.com?utm_medium=email&utm_source=footer
> [2] https://groups.google.com/d/optout