Re: [qubes-users] Re: Best works-out-of-the-box dual-head gfx card?
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) -- 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+unsubscr...@googlegroups.com. To post to this group, send email to qubes-users@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/qubes-users/a41e44b3-eaf2-2739-3df3-e46f5fc6c7c5%40gmx.com. For more options, visit https://groups.google.com/d/optout.
Re: [qubes-users] Re: Best works-out-of-the-box dual-head gfx card?
On Friday, March 9, 2018 at 1:04:53 AM UTC+1, tai...@gmx.com wrote: > 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. 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? -- 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+unsubscr...@googlegroups.com. To post to this group, send email to qubes-users@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/qubes-users/c0315dc9-dc9e-4ca4-8abb-632677db9f71%40googlegroups.com. For more options, visit https://groups.google.com/d/optout.
Re: [qubes-users] Re: Best works-out-of-the-box dual-head gfx card?
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. -- 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+unsubscr...@googlegroups.com. To post to this group, send email to qubes-users@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/qubes-users/1e867115-e282-d5f8-a977-af77fd112415%40gmx.com. For more options, visit https://groups.google.com/d/optout.
[qubes-users] Re: Best works-out-of-the-box dual-head gfx card?
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 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! 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). -- 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+unsubscr...@googlegroups.com. To post to this group, send email to qubes-users@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/qubes-users/950fa0d8-aa88-4af4-a6ff-81f092db293c%40googlegroups.com. For more options, visit https://groups.google.com/d/optout.
[qubes-users] Re: Best works-out-of-the-box dual-head gfx card?
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
