I really hope , the developers of the multiseat software .. enable us to use 1 gpu for multiple users/seats , with warnings or what have you . The Aster multiseat , already does that , with a more user friendly interface , seems much easier to use , with customizability on the go . Not to discredit the devs of this here and multiseat x , its amazing as it is .
Yeah same here, I just want to allow my wife to search your sites while I game. I have windows 11 using aster. That works fine. But now I'm looking at using linux more I was hoping to configure the same. Aster licensing soon counters the savings after a couple of years of using it. I have been comparing windows 11 vs linux gaming as well. There is not much difference at setting up games on steam, and they all run well so far.
Those guys(one in particular hahaha) got all my respect. You see, i'm looking for something like that for a life... and only when i was introduced to the keyword "multi seat" that i find my self in this confy place. I will try this setup tomorrow in Ubuntu 22.04 and hope that evething works just fine. But before all of that, thanks you, really, for all the efforts and knowledges that you have shared here right now.
@msscash how did you get on? I've been planning this for some time. I'm a Linux novice so hoping to learn on the job. I'm hoping to use scavenged hardware (core2 quad era type stuff) the office was throwing out. Any idea how strict the one graphics adapter per user is? The GPUs (Nvidia quadro type things) I have have at least two outputs. Ideally I want 3-4 seats, not too concerned with gaming. Hopefully it's possible with my collection of GPU 🤞
I set up a couple of windows “multi seat” systems about 10 years ago - you could game on main system but the virtual seats were software rendered only, so you could barely watch youtube. This looks way more capable and much easier!
It's very capable, each GPU goes entirely to one seat and everything else is shared. To go to >1 user per GPU you do lose hardware rendering on the additional seats though. In my case seat1 got the RS580 so it's more powerful than the first seat (seat0).
I've been trying to get single display multiseat working for a long time, so I'm really interesting if you can can get it working. Love the current setup, it connected a lot of the dots that I've been trying to do on one of my setups. So thanks for that!!
Single GPU you mean? Presumably to have a seat you need separate displays lol But yeah, it's amazing how out of date all of the tutorials are on this, yet it's actually a rather solid implementation by everything involved (esp. systemd and X).
@@apalrdsadventures , single GPU multiseat implemented via Xephyr can be used regardless of the number of displays. For a single display with multiple seats, just have two Xephyr windows on the one display.
I think you should consider re-making this video with a guide for a single-gpu system, I remember it was possible, or at least seemed to be a few years ago. This vid. gives people the wrong idea that you have to have a second GPU in the system. Also, there is yet another way of setting this up - with containers (LXC/LXD) or systemd system slices - I think in this case you'd need a separate GPU tho. This way has the plus of giving each use a separate OS userland running on the same, shared kernel. btw. please keep in mind that I'm talking about OS containers, not application containers (docker for example). People these days are notriously confused as to what docker/containers/cgroups actually is and does.
I really enjoyed your video and competent guidance. I had the same level of interest in multi-seat starting on linux pre-systemd! I had limited success on the cli. Then with Userful from Canada I had full success. Nowadays systemd makes it so much easier. Currently i run FatDog64 without systemd. Its still possible to pass hardware through directly to a virtual OS, as long as the kernel has been compiled with kvm, iommu or VT-d, passthrough. Qemu also allows partitioning cpu threads and memory to each virtual OS. FatDog, strangely, also runs faster as a virtual OS than it does on bare metal.
Great video! 👍 Are there any news about multi-seat on a single (integrated) GPU? Having to use multiple dedicated GPUs (one per seat) on an office machine is quite a power hog (e.g. 3x GPU for 3 users). It's a pity Aster can do this on Windows and we don't get it working on Linux. 😞
You can get pretty cheap low end workstation GPUs (PCIe slot powered), but yeah it would be nice to do it not one GPU per seat. I haven't had a chance to work on this project in awhile unfortunately
@@apalrdsadventures Aside from the power consumption the number of PCIe slots is a big limitation for multi-seat systems with exclusive GPUs for users.
Год назад
Excelente video. I was there, in the 90s also wondering what kind of magic what that. I managed to have a couple of seats working but was very inconsistent. One caveat: you might need to add your users to group video, and some other supplemental groups.
I have a little fast NUC with two display connections. They both work flawlessly. Still - just one graphics card, but I wouldn't need acceleration or would be happy to have acceleration on just one account. I would love to use it as a two-seat PC for normal office stuff. My NUC has 32GB RAM and is super fast. So performance should be no issue. It's almost incredible, that there are so few articles, videos on multiseat. I would have expected to find a distri specifically tailored to multiseat with everything installed and a GUI for configuration. But NOTHING! I feel like this should be supported more by the Linux community.
@@apalrdsadventures all proxmox related posts. left my hyperv server behind, all running on proxmox now.!Love the port forwarding part of it. Now running my domoticz virtual.
the systemd developers can't think of a situation where 2 users seated side by side looking at each others keyboards might decide to just trust each others graphics buffers. Why? Because they have never been that tech poor. There are ways to share a GPU even within systemd using xnest but they don't make it easy to restrict the HID to each session. Given that the GPU when gaming can easily cost more than the rest of the system this is a serious problem for poorer countries and schools, but not for the tech bros
For schools, you live with the lack of hardware acceleration and nest. You should still get hardware transcode (libva/vaapi) but not OpenGL/Vulkan, since transcode is already shareable. For gaming though, it's not as simple as the GPU buffers are shared, sharing is bad, although that's the reason kernel devs have given. The entire DRM pipeline (Direct Rendering Manager) basically treats multiple output cards as a single card with a really big output canvas, and the X server then knows the coordinates of each output. It's a more thorough design issue with DRM operating at the card level instead of the output level. Hence the sysfs refering to card0 and renderD128. SystemD for their part is aware of the issues and has promised to fix anything they can on their end whenever DRM supports separating the card outputs into separately controllable devices. Currently the individual output devices in sysfs are purely for edid / resolution control and absolutely all rendering goes through the shared node, so standing up two X servers means they can't both share the same card0 node and systemd would have to nest or proxy the render commands from both users into one.
Indeed, the desire to have multiple seats with _one GPU_ is pretty much the only reason I'm interested in a multi-seat setup. I should be experimenting with Xephyr soon to see if I can get that working.
@@JivanPal we should talk about it then. I did a lot of work on it a few months ago but was not ready to implement it. I came up with 3 different options. The real problem is isolating users so that one cannot shutdown the machine when ending their session. And also to seperate the HID inputs which has a lot of ugly ways to set up. Apalrd is wrong BTW. It was not stopped in the kernel but in systemd after a dependancy was no longer supported.
@@kimcosmos Looks like RUclips hid my previous reply to you in which I described what I plan to do henceforth, including testing Xephyr. Well, after some digging, it seems that Xorg removed support for evdev, so there is no officially supported way to conveniently isolate input devices to specific Xephyr windows. Someone made a PPA that patched Xorg to restore this functionality, but that was back when Ubuntu 16.04 was current; it now remans unmaintained. I will have to see if I can compile the latest stable version of Xorg with the patch. If I can do that, we should be in business.
Lots of people say bad things about data deduplication being a huge speed performance penalty and RAM resource hit in ZFS. It looks fine to me from the video but was there anything behind the scenes that you had difficulty with regarding deduplication?
Dedup in zfs definitely does use some RAM for the dedup tables. It's possible to improve this somewhat on large setups by using dedicated SSDs for the dedup tables. In a client system like this with two users, they aren't hitting the drive hard enough continuously to care and if they are gaming together it's likely that they will be loading similar assets together so caching performance will be better.
@@apalrdsadventures That's what I thought about which given what I've read from Googling and the various things on Reddit. Makes me feel like people are blind to compression adjacent techniques because stuff is often compressed already and it's not worth it. Years ago when I messed around with it on the Windows Server 2012r2 chunk level and more recently the file level deduplication that exists for APFS in Mac OS, it's surprising that on the small scale of one household or user that it will still find and save a meaningful amount.
A while ago I did that with Borderlands 2. In that case I just force one user on controller and the other on mouse/keyboard, launching the instances on separate displays. And with a single GPU. Much easier when the game can accept input while not in focus.
Nice video! I've read somewhere that you could create "virtual GPUs" on Linux with the professional GPUs from nvidia and AMD (Radeon Pro), which could be used to have a multiseat setup with hardware acceleration on all seats while only using one real graphics card. Is that true? This would be awesome! But I can't find much information on this topic online... I just wish one day this will be possible to do with a mainstream consumer GPU.
AFAIK it's not that easy. Normally GPU partitioning is done to feed the partitions to multiple virtual machines, so it's designed for cards without any graphic outputs. With nVidia, there are some scripts that make the drivers work on Geforce cards with the same silicon as their professional equivalents, but you still lose the graphic outputs. I'm not sure if anyone has tried to partition a GPU and then use it in the same Linux system hosting the physical GPU.
the setting up is really painful. i hoped it just so easy to do, just like Aster in Windows. just drag and drop devices to the second desktop. it also notifies us, which devices is currently pressed/moved cursor for mouse, so we could easily identify which devices needed to be dragged. Aster could expand easily to 6 desktop, as compared to this method would be painful as hell for more than 2 seats.
This is extremely useful solution for couch gaming setup. While one user is playing on couch, desktop is totally free to use. How ever I would like to do this with shared GPU so I don't HAVE TO play on couch but I can play also from desktop.
Small addition: You mentioned the VM-solution only being able to use 50% of the resources on each VM. With dynamically allocated resources, this is not the case.
CPU resources can absolutely by allocated dynamically. AFAIK you can't use kvm ballooning + pcie passthrough at the same time, so you're limited on the memory side if you are doing GPU passthrough. I haven't found any more recent info suggesting it's become possible in the last year or so.
@@apalrdsadventures I just was here to say the same. RAM balloning is not very well implemented in all guests systems. the only thing you can do is KSM (kernel same page merging) but that is another feature. With KSM enabled you could potentially run more VM than RAM you have due to KSM+Copy on Write feature. KSM is disabled by default because of similar security problems that spectre have.
correct me if I'm wrong but from what I am understanding, right now you have 2 users using one MB but require separate graphics units, such as the GPU and the APU. Theoretically then, as long as you have have the same number of GPUs for users on the same computer, you can have more then just 2 users, correct?
This was really cool I'm gonna be sure trying it out and I can't wait till the follow-up. I'm going to look into if it's possible to have multiseat in a split screen configuration with virtual displays say for a living room PC and multiple controllers
Yes it's totally possible, but when you use one GPU for multiple displays you have to nest X servers and lose OpenGL (and the documentation on it all predates Vulkan). When you use each graphics output on one GPU for a different user you basically make a giant desktop the size of all of the monitors, then run each user session in a full-screen 'window' with the coordinates of one of the monitors. So nothing would stop you from making a split screen quad instead of using multiple monitors, except the games won't run well since there's no hardware acceleration
Looking to invest in a CLX but have no idea how to turn it into a multi seat. Now listening to these guys I can request what I need in the build. If it weren't for people like these guys , computer makers will never tell you how to do multiseat because sales would slump.
I've tried it on my pc (5600G and RX6600), works well on X11 but couldn't get it to work on wayland for some reason, I also found that seat0 which has de igpu can still use the discrete gpu for rendering by using DRI_PRIME=1, if it gives you an error you have to add your user to the 'render' group, so basically both users could share the more powerful gpu for rendering their games, you should try to limit fps on the games that you run so that they don't take 100% of the gpu, not sure how the priority of rendering works but playing the same game on both seats usually results in similar performance for both but sometimes it varies. Pretty neat!
Oh, something pretty weird, or maybe not idk, I can actually use wayland but only on one user, if i try to launch the second seat using wayland too i get a black screen and an X pointer, and after a while it takes me back to sddm (it also freezes the the wayland seat for a while, but hitting ctrl+alt+f1 and then switching back resumes it), so I use wayland on one seat and then x11 on the other. seems weird but somehow works so xd
Also, for steam, I made a /games partition (btrfs subvolume actually but anyway) and added the steam library to both users, not sure if it could conflict or something, maybe if there are updates, but haven't encountered issues yet(?)
Excellent video! I'm gonna try something similar to this for a local game server, but I have a question: Does each seat get it's own IP address on the network? Or does some kind of subnetting go on? I'm gonna try to make a "one machine LAN party" setup, are there things I'll need to know about going into it?
Both users are logged in to the same system at the same time, so one IP for the system similar to a system with a lot of ssh sessions. You can add multiple IPs to the system though, but then the app needs to choose which to bind to. You can also use network namespaces (netns) to give a proper separate networking set for an app, although that's a bit more involved.
I have a question on this. Can you take the solution - apply sunshine/moonlight and make a cloud gaming multi-seat server. I have seen people do that with VM's splitting the GPU(s) into the VMs, but perhaps this solution offers a simpler road to the same end.
Reading their docs it might be possible, but it would need to be setup with two separate daemons on separate ports (or two virtual IPs) and with two separate config files. It does have options to select the GPU to capture and monitor on the GPU (Sunshine/Moonlight are limited to 1 display due to the nvidia protocol), so it seems like it should be able to work. I have not tested it though. It's of course limited to 1 session per GPU and Linux only, but Linux gaming has come pretty far.
Thank you very much for this tutorial, apalrd : clear, simple and efficient. I set up the two desktops configuration I needed at home (Ubuntu 22.04 + Intel motherboard + Amd Radeon RX6600) and it worked like a charm (Wayland and/or Xorg). My only trouble is with seat1 : I can't leave or change a user session without totally crashing it + starting over a user session on seat0. Not very handy as a desktop configuration. Is it the same for you and, if so, do you have a clue why ?
3:40.. Is it MoBo "ASRock B450M Pro4"? Does the motherboard support IOMMU? 9:50.. With this iGPU, does the following file exist? File path: ls "/sys/bus/pci/devices/0000:07:00.0/rom" I ask because I am looking for hardware for a low cost homelab with headless hypervisor with single GPU passthrough.
Yeah it's the B450M Pro4. IOMMU setting is in Advanced->AMD CBS->NBIO Common Options->NB Configuration and defaults to disabled (of course) but once that's fixed it does support IOMMU in Linux. I wouldn't say I picked it because it was a good choice, it was low cost at the time and fit my needs. I was using the iGPU for transcode for my security system at the time, but that was in an LXC container so I wasn't passing it through. I've successfully passed through a GPU in the primary PCIe slot in Proxmox previously on this hardware. At that point I was using the Vega GPU as the BIOS GPU and also for Proxmox, so the second GPU was uninitialized. Looking at the IOMMU groups, it looks like at best the Vega GPU will only pass through without HDMI audio (since the USB root, PSP, and HD Audio all share a group with the HDMI audio controller, but the GPU itself is in its own group).
9:12 Ah yeah, there goes HDR haha Wayland has been taking its sweet sweet time reaching feature parity (it's getting there!) but now with broader drm_lease compatibility (gnome merged protocol support just recently) I feel like that'd be ideal for sharing a single video card via logind multiseat? Possibly... I haven't given it a try - I'm mostly spitballing after seeing this video. I'm sure someone out there knows why that won't work. Like maybe compositors have yet to do some funny mode setting shenanigans? I'd love to know either way.
@@apalrdsadventures Thank you. I was just hoping that this would be possible on a single GPU. I have done the weird GPU-PV thing that some youtubers tried a while back, but that has been abandoned and the new implementation by Microsoft just ignores AMD GPUs... p.s.: Found your blog! VERY HELPFUL!
If open-source GPU drivers from AMD would allow resource sharing between multiple users then this solution would be interesting. Of course I'm skipping GPUs from NV because in this case it's unrealistic. (closed drivers)
There's very slight evidence that AMD is currently working on this, at least a way for multiple users to submit work to the GPU. I assume it's for shared compute / OpenCL reasons, but there is kernel work going on to amdgpu to deal with multiple users. NV does their own whole mess on Linux
Very nice tutorial thank you so much :-) For now I think I will stick with my pci passthrough but when I get into games that require non vm's I may try this solution.
i use aster multiseat and sandboxie plus application copier, works very nice with Minecraft once u get use to the setup wich adds 5 or so mins with turing stuff on and all that
Thanks for the video, I needed it not for gaming but because my wife's computer broke and we didn't want to buy another. Ubuntu wouldn't work out of the box and I'm not good with computers, but Rocky Linux worked right away
Interesting! Just to mention, u can do gpu passthrough with qemu/kvm without wasting one gpu for the host. I do this with proxmox all time. Have 2 gpus plus igpu, run linux vm on igpu for proxmox webpanel, 2 additional vms on the gpus.
I don't have any Bluetooth hardware to try this with, unfortunately. Do they show up as separate devices in seat-status? Worst case, it's certainly possible to split them up if they are on USB or use their own USB dongles instead of Bluetooth.
i've got multiseat working but i want to share it via the internet, there's this app called sunshine but i'm not able to assign the virtual devices that it creates to the 2nd seat
Bummer. No amd and nvidia mix option !? I really need one monitor multi seating. I tried Hyper-v with gpu partitioning("ruclips.net/video/XLLcc29EZ_8/видео.html") and worked but I cant get any way to multi seat with it. Tried every 3rd party pay option for mouse. Nun work as I cant see the curser. It's there just cant see it. :( Any suggestions ?
Blame nvidia for messing with the gl / vk libraries instead of using the proper Linux display system, as to why multi seat doesn't work. What are you trying to do with one monitor? Or do you mean one GPU?
@@apalrdsadventures I'm trying to get 3 gaming instances on 2 monitors. but cant get moues/keyboards to work to its own desktop. Vmwear doesn't have the correct direct-x :( But it does allow me to have multi mouse and keyboard to each vm.
It's the same SystemD API that I'm using here, Wayland should correctly implement it but I found it didn't work correctly for OpenGL and especially Vulkan.
@@apalrdsadventures Technically Vulkan can do that with offscreen framebuffer, but the only problem is how programmers understand it (personally did I give up to code it last project). I don't know about OpenGL since Wayland only uses EGL for every OpenGL operations
With Wayland I had Vulkan rendering to the default GPU (the Raven Ridge) off screen and it was around half the speed the GPU should have been capable of (presumably due to the system memory copy of the framebuffer from one GPU to another), even without a game running on the Raven GPU. With X the games were correctly running on the GPU for their seat. I'm assuming X is setting the mesa vk preferred GPU environment variable and Wayland is not, or it's not setting it differently per seat.
The little keyboard+touchpad is a GearHead Smart Touch II, but it's not sold any more. The nicer ones are a pair of Razer keyboard+mouse (they are my brothers), and I use a Unicomp IBM replica and Logitech MX Verticla mouse.
Very game dependent. Some games also tend to be CPU bound because they are single-threaded or lightly multi-threaded, so having 8 CPU threads in this setup can deal with 2 CPU bound games which only run on 2-3 threads each.
My friends and I played 4 players in the game aliens: colonial marines on one RTX2060 Super 8Gb video card. Windows 10 + ASTER will be cheaper for 4 players.
It's cheaper to get a pci-e x1 to x16 adapter and a 9600 gt than the cheapest license of aster (which is only for 2 seats). Which is good enough for games like xonotic, 0ad, Warzone2100, and maybe even Portal. If you want to be able to play something more resource intensive, yes, aster is the way to go. I couldn't find a crack that works for aster, if you know one let me know. I don't speak a lot of english, sorry if my comment has lots of grammar errors and stuff.
I got a headache trying to setup a multi-sest between amd r9 380 and nvidia 8600gts and couldn't make it go past the ubuntu shows on all displays and had only multiseat in text... so I bought aster 3 seats and am stuck with a windows 7 multiseat that works, only issue is gpu instability between amd/nvidia resulting in random reboots but disabling hardware acceleration in firefox about:config resolves that
My setup is 1 seat with amd r9 380, I game on that one, and 2 seats on nvidia 8600gts, one on svideo 720x480 32" panasonic crt tv, running kodi, the other a 40" hdtv in 1080p running kodi and/or firefox. System can game on amd r9 and do 2 tv entertainment systems, i can run old emulators inside kodi that is fun and i have no problems running 2 kodi sessions on that old generation nvidia gpu. All Firefox gpu acceleration is disabled.
Sorta. I was able to use one account to play singleplayer, but it wouldn't sync my save games to both at the same time. So one seat would end up starting from the beginning. You also can't play multiplayer with your own account, so no LAN parties or co-op if you don't have separate accounts.
@@apalrdsadventures I was just able to finish this video. I am very grateful you made it. I have kids and tried this myself for homework stations, but couldn't get sound to work at the second station. Or it was first come first serve with audio. A very frustrating experience. I'm going to try this again soon using your video as a guide. Thank you again!!
Glad you enjoyed it! My next goal for this is to use DisplayLink adapters (USB3 laptop docks with USB3 graphics devices) which should bring the cost down a lot if you don't need 3D acceleration
i wish there was, it was a pain for me to try multiseats and didn't work for me, i had to purchase a licence of aster for windows 7 which is now not supported anymore, i wish it would work on linux the easy way too
Yes. Too bad taht whenever we start to get all the kinks ironed from what is used (xorg) they have to be switched up because of security (wayland) forcing everything to lose 10 years of development every switch. 20 years of evolution, 10 years of devolution, always like that.
Not directly If you're using a hypervisor, you would pass through the entire PCIe device using vfio instead of a method like this. The GPU would be bound to the VM, and the VM would control it completely
Not if you use PCIe passthrough - at least not using this method. You'd still have a multi-seat system, but via virtual machines which partitioned access to the hardware.
It already kinda works, but gdm gets frozen when logging in on the second seat. Also, I think gnome-shell refuses to run in second wayland session, but you can still run another compositor like weston or even kde plasma
It's more of a 'Microsoft doesn't want to let you' than 'it's technically impossible' with Windows, since it absolutely supports multiple user sessions over RDP (with a server license of course).
you can if you install aster and maybe buy a 2 or 3 seat licence which costs less than $100. I run it with a amd r9 380 and an old nvidia 8800 gts for a retro pc with svideo tv out
You should have started MUCH simpler. Just one graphics card - maybe even one of those Intel ones. Either NO acceleration for BOTH seats or acceleration for seat0 and no acceleration for seat1. Would this be possible? Much easier? How?
It's much easier to split a system up by GPUs than to split a single GPU into multiple seats, mostly due to the architecture of the Linux DRM/DRI being card-based and not port-based. Each card needs to be 'owned' by an X server, which runs in the user session. Any app which wants to perform accelerated rendering needs to go through the X server for authorization to access the card's render functions. For two user sessions to share a card, you need to spawn a third global session and forward the rendered output of each user session, and the user sessions wouldn't get acceleration access. This is a much more complicated setup than just having two cards = two X servers = two users.
Hey guys . I've some suggestions . Why don't you guys try WSL ( Windows subsystem linux )version 2 . I don't assume you guys use Windows , but this is too awesome to pass up . It lets you install a linux distribution , with way too much cooperation with windows , such that you can drag linux windows through windows ! it shares the hypervisor of the windows , and doesn't stack or nest hypervisors or kernels , but the linux and windows kernel works alongside each other . I want to see if you can multiseat using a windows software but through that , run linux software with just one gpu .
@@joelpichette its a little different than trying linux on windows . In some ways better even . It kinda works like wine ( acronym for ' wine is not an emulator ) . Helps run windows software on linux .
Very good video. No quick cuts, calm explanations. ++
Thanks!
I really hope , the developers of the multiseat software .. enable us to use 1 gpu for multiple users/seats , with warnings or what have you .
The Aster multiseat , already does that , with a more user friendly interface , seems much easier to use , with customizability on the go .
Not to discredit the devs of this here and multiseat x , its amazing as it is .
Yes, exactly that is what i wanted to say.
Yeah same here, I just want to allow my wife to search your sites while I game. I have windows 11 using aster. That works fine. But now I'm looking at using linux more I was hoping to configure the same. Aster licensing soon counters the savings after a couple of years of using it.
I have been comparing windows 11 vs linux gaming as well. There is not much difference at setting up games on steam, and they all run well so far.
Those guys(one in particular hahaha) got all my respect. You see, i'm looking for something like that for a life... and only when i was introduced to the keyword "multi seat" that i find my self in this confy place. I will try this setup tomorrow in Ubuntu 22.04 and hope that evething works just fine. But before all of that, thanks you, really, for all the efforts and knowledges that you have shared here right now.
@msscash how did you get on?
I've been planning this for some time. I'm a Linux novice so hoping to learn on the job.
I'm hoping to use scavenged hardware (core2 quad era type stuff) the office was throwing out. Any idea how strict the one graphics adapter per user is?
The GPUs (Nvidia quadro type things) I have have at least two outputs. Ideally I want 3-4 seats, not too concerned with gaming. Hopefully it's possible with my collection of GPU 🤞
I set up a couple of windows “multi seat” systems about 10 years ago - you could game on main system but the virtual seats were software rendered only, so you could barely watch youtube. This looks way more capable and much easier!
It's very capable, each GPU goes entirely to one seat and everything else is shared. To go to >1 user per GPU you do lose hardware rendering on the additional seats though. In my case seat1 got the RS580 so it's more powerful than the first seat (seat0).
@@apalrdsadventuresAFAIK wayland has drm lease extension that allows multiseat with hardware rendering by letting each seat to have own compositor.
I've been trying to get single display multiseat working for a long time, so I'm really interesting if you can can get it working. Love the current setup, it connected a lot of the dots that I've been trying to do on one of my setups. So thanks for that!!
Single GPU you mean? Presumably to have a seat you need separate displays lol
But yeah, it's amazing how out of date all of the tutorials are on this, yet it's actually a rather solid implementation by everything involved (esp. systemd and X).
In a way , you can through the use of capture card ...
@@apalrdsadventures , single GPU multiseat implemented via Xephyr can be used regardless of the number of displays. For a single display with multiple seats, just have two Xephyr windows on the one display.
I think you should consider re-making this video with a guide for a single-gpu system, I remember it was possible, or at least seemed to be a few years ago. This vid. gives people the wrong idea that you have to have a second GPU in the system.
Also, there is yet another way of setting this up - with containers (LXC/LXD) or systemd system slices - I think in this case you'd need a separate GPU tho. This way has the plus of giving each use a separate OS userland running on the same, shared kernel.
btw. please keep in mind that I'm talking about OS containers, not application containers (docker for example). People these days are notriously confused as to what docker/containers/cgroups actually is and does.
It's possible but not nearly as easy as you make it seem, especially if you want good 3d acceleration for both OpenGL and Vulkan.
Yes please I want to achieve it as well, I am even considering multi cursor at this point
I really enjoyed your video and competent guidance. I had the same level of interest in multi-seat starting on linux pre-systemd! I had limited success on the cli. Then with Userful from Canada I had full success. Nowadays systemd makes it so much easier. Currently i run FatDog64 without systemd. Its still possible to pass hardware through directly to a virtual OS, as long as the kernel has been compiled with kvm, iommu or VT-d, passthrough. Qemu also allows partitioning cpu threads and memory to each virtual OS. FatDog, strangely, also runs faster as a virtual OS than it does on bare metal.
Great video! 👍
Are there any news about multi-seat on a single (integrated) GPU?
Having to use multiple dedicated GPUs (one per seat) on an office machine is quite a power hog (e.g. 3x GPU for 3 users).
It's a pity Aster can do this on Windows and we don't get it working on Linux. 😞
You can get pretty cheap low end workstation GPUs (PCIe slot powered), but yeah it would be nice to do it not one GPU per seat.
I haven't had a chance to work on this project in awhile unfortunately
@@apalrdsadventures Aside from the power consumption the number of PCIe slots is a big limitation for multi-seat systems with exclusive GPUs for users.
Excelente video. I was there, in the 90s also wondering what kind of magic what that. I managed to have a couple of seats working but was very inconsistent.
One caveat: you might need to add your users to group video, and some other supplemental groups.
I have a little fast NUC with two display connections. They both work flawlessly. Still - just one graphics card, but I wouldn't need acceleration or would be happy to have acceleration on just one account.
I would love to use it as a two-seat PC for normal office stuff. My NUC has 32GB RAM and is super fast. So performance should be no issue.
It's almost incredible, that there are so few articles, videos on multiseat. I would have expected to find a distri specifically tailored to multiseat with everything installed and a GUI for configuration. But NOTHING!
I feel like this should be supported more by the Linux community.
superb. Can't wait for your next video. i've binged all the old ones....
Which one was your favorite?
@@apalrdsadventures all proxmox related posts. left my hyperv server behind, all running on proxmox now.!Love the port forwarding part of it. Now running my domoticz virtual.
Well you're in luck, Proxmox is definitely well represented on the ideas list
the systemd developers can't think of a situation where 2 users seated side by side looking at each others keyboards might decide to just trust each others graphics buffers. Why? Because they have never been that tech poor. There are ways to share a GPU even within systemd using xnest but they don't make it easy to restrict the HID to each session. Given that the GPU when gaming can easily cost more than the rest of the system this is a serious problem for poorer countries and schools, but not for the tech bros
For schools, you live with the lack of hardware acceleration and nest. You should still get hardware transcode (libva/vaapi) but not OpenGL/Vulkan, since transcode is already shareable.
For gaming though, it's not as simple as the GPU buffers are shared, sharing is bad, although that's the reason kernel devs have given. The entire DRM pipeline (Direct Rendering Manager) basically treats multiple output cards as a single card with a really big output canvas, and the X server then knows the coordinates of each output. It's a more thorough design issue with DRM operating at the card level instead of the output level. Hence the sysfs refering to card0 and renderD128.
SystemD for their part is aware of the issues and has promised to fix anything they can on their end whenever DRM supports separating the card outputs into separately controllable devices. Currently the individual output devices in sysfs are purely for edid / resolution control and absolutely all rendering goes through the shared node, so standing up two X servers means they can't both share the same card0 node and systemd would have to nest or proxy the render commands from both users into one.
Indeed, the desire to have multiple seats with _one GPU_ is pretty much the only reason I'm interested in a multi-seat setup. I should be experimenting with Xephyr soon to see if I can get that working.
@@JivanPal we should talk about it then. I did a lot of work on it a few months ago but was not ready to implement it. I came up with 3 different options. The real problem is isolating users so that one cannot shutdown the machine when ending their session. And also to seperate the HID inputs which has a lot of ugly ways to set up. Apalrd is wrong BTW. It was not stopped in the kernel but in systemd after a dependancy was no longer supported.
@@kimcosmos Looks like RUclips hid my previous reply to you in which I described what I plan to do henceforth, including testing Xephyr. Well, after some digging, it seems that Xorg removed support for evdev, so there is no officially supported way to conveniently isolate input devices to specific Xephyr windows. Someone made a PPA that patched Xorg to restore this functionality, but that was back when Ubuntu 16.04 was current; it now remans unmaintained. I will have to see if I can compile the latest stable version of Xorg with the patch. If I can do that, we should be in business.
@@JivanPal I came up with a few ways but none great. Do you use your twitter? I don't really
Is there a way to make two seats on one video card just for general work (video, music and youtube) not for gaming?
Kinda sorta. There are older tutorials that use x nesting to achieve this, but none are really recent.
Great video! It would be awesome to see you re-visit this setup with Wayland at some point in the future.
Eventually, if Wayland figures their stuff out and properly supports everything.
Figure out what? The Wayland protocol is purposefully spartan. It serves very specific use cases that don't include workstations.
Loved every second. Your explanation is easy to understand. Though I won't do this, it's fun to watch. Subbed...
Glad you enjoyed it!
Good video, nicely explained. I've subbed, good stuff :) Strange you guys don't have more subs.
Great work .... Loved it. Waiting for "share graphics output on a single graphics". ...
Lots of people say bad things about data deduplication being a huge speed performance penalty and RAM resource hit in ZFS. It looks fine to me from the video but was there anything behind the scenes that you had difficulty with regarding deduplication?
Dedup in zfs definitely does use some RAM for the dedup tables. It's possible to improve this somewhat on large setups by using dedicated SSDs for the dedup tables.
In a client system like this with two users, they aren't hitting the drive hard enough continuously to care and if they are gaming together it's likely that they will be loading similar assets together so caching performance will be better.
@@apalrdsadventures That's what I thought about which given what I've read from Googling and the various things on Reddit. Makes me feel like people are blind to compression adjacent techniques because stuff is often compressed already and it's not worth it. Years ago when I messed around with it on the Windows Server 2012r2 chunk level and more recently the file level deduplication that exists for APFS in Mac OS, it's surprising that on the small scale of one household or user that it will still find and save a meaningful amount.
A while ago I did that with Borderlands 2.
In that case I just force one user on controller and the other on mouse/keyboard, launching the instances on separate displays. And with a single GPU.
Much easier when the game can accept input while not in focus.
How about wayland an Multiseat now? How about 23.10? Is there a system which runs more smoothly concerning proceesor Power?
Nice video! I've read somewhere that you could create "virtual GPUs" on Linux with the professional GPUs from nvidia and AMD (Radeon Pro), which could be used to have a multiseat setup with hardware acceleration on all seats while only using one real graphics card. Is that true? This would be awesome! But I can't find much information on this topic online...
I just wish one day this will be possible to do with a mainstream consumer GPU.
AFAIK it's not that easy. Normally GPU partitioning is done to feed the partitions to multiple virtual machines, so it's designed for cards without any graphic outputs. With nVidia, there are some scripts that make the drivers work on Geforce cards with the same silicon as their professional equivalents, but you still lose the graphic outputs.
I'm not sure if anyone has tried to partition a GPU and then use it in the same Linux system hosting the physical GPU.
On linux all consumer-grade gpus can do it. Probably only nvidia shit driver can't do it.
@@uis246 So how can I create two virtual GPUs (that are both capable of 3D acceleration) on my single AMD GPU?
the setting up is really painful. i hoped it just so easy to do, just like Aster in Windows. just drag and drop devices to the second desktop. it also notifies us, which devices is currently pressed/moved cursor for mouse, so we could easily identify which devices needed to be dragged. Aster could expand easily to 6 desktop, as compared to this method would be painful as hell for more than 2 seats.
I'm also on a journey for multiseat, thank you for sharing this video😁 have you tried virtualgl+thinklinc server?
I haven't tried VirtualGL yet, but it's on my radar for the future
This is extremely useful solution for couch gaming setup.
While one user is playing on couch, desktop is totally free to use.
How ever I would like to do this with shared GPU so I don't HAVE TO play on couch but I can play also from desktop.
Small addition:
You mentioned the VM-solution only being able to use 50% of the resources on each VM. With dynamically allocated resources, this is not the case.
CPU resources can absolutely by allocated dynamically.
AFAIK you can't use kvm ballooning + pcie passthrough at the same time, so you're limited on the memory side if you are doing GPU passthrough. I haven't found any more recent info suggesting it's become possible in the last year or so.
@@apalrdsadventures I just was here to say the same.
RAM balloning is not very well implemented in all guests systems.
the only thing you can do is KSM (kernel same page merging) but that is another feature.
With KSM enabled you could potentially run more VM than RAM you have due to KSM+Copy on Write feature.
KSM is disabled by default because of similar security problems that spectre have.
correct me if I'm wrong but from what I am understanding, right now you have 2 users using one MB but require separate graphics units, such as the GPU and the APU.
Theoretically then, as long as you have have the same number of GPUs for users on the same computer, you can have more then just 2 users, correct?
Yes, that does work. There are some limits on mixing GPUs from different vendors, especially nvidia, but in my case all-AMD works fine.
This was really cool I'm gonna be sure trying it out and I can't wait till the follow-up. I'm going to look into if it's possible to have multiseat in a split screen configuration with virtual displays say for a living room PC and multiple controllers
Yes it's totally possible, but when you use one GPU for multiple displays you have to nest X servers and lose OpenGL (and the documentation on it all predates Vulkan).
When you use each graphics output on one GPU for a different user you basically make a giant desktop the size of all of the monitors, then run each user session in a full-screen 'window' with the coordinates of one of the monitors. So nothing would stop you from making a split screen quad instead of using multiple monitors, except the games won't run well since there's no hardware acceleration
Looking to invest in a CLX but have no idea how to turn it into a multi seat. Now listening to these guys I can request what I need in the build. If it weren't for people like these guys , computer makers will never tell you how to do multiseat because sales would slump.
Best video of explaining Aster for gaming. i liked and subscribed
I've tried it on my pc (5600G and RX6600), works well on X11 but couldn't get it to work on wayland for some reason, I also found that seat0 which has de igpu can still use the discrete gpu for rendering by using DRI_PRIME=1, if it gives you an error you have to add your user to the 'render' group, so basically both users could share the more powerful gpu for rendering their games, you should try to limit fps on the games that you run so that they don't take 100% of the gpu, not sure how the priority of rendering works but playing the same game on both seats usually results in similar performance for both but sometimes it varies. Pretty neat!
Oh, something pretty weird, or maybe not idk, I can actually use wayland but only on one user, if i try to launch the second seat using wayland too i get a black screen and an X pointer, and after a while it takes me back to sddm (it also freezes the the wayland seat for a while, but hitting ctrl+alt+f1 and then switching back resumes it), so I use wayland on one seat and then x11 on the other. seems weird but somehow works so xd
Also, for steam, I made a /games partition (btrfs subvolume actually but anyway) and added the steam library to both users, not sure if it could conflict or something, maybe if there are updates, but haven't encountered issues yet(?)
Could you make an updated version for this video?
Excellent video! I'm gonna try something similar to this for a local game server, but I have a question: Does each seat get it's own IP address on the network? Or does some kind of subnetting go on?
I'm gonna try to make a "one machine LAN party" setup, are there things I'll need to know about going into it?
Both users are logged in to the same system at the same time, so one IP for the system similar to a system with a lot of ssh sessions. You can add multiple IPs to the system though, but then the app needs to choose which to bind to. You can also use network namespaces (netns) to give a proper separate networking set for an app, although that's a bit more involved.
@@apalrdsadventures ah, so y'all are playing port-to-port on the same IP. Thank you!
What software do you use to make your RUclips videos?
Mostly Davinci Resolve
This is awesome! Thanks for sharing!
Glad you enjoyed it!
I have a question on this. Can you take the solution - apply sunshine/moonlight and make a cloud gaming multi-seat server. I have seen people do that with VM's splitting the GPU(s) into the VMs, but perhaps this solution offers a simpler road to the same end.
Reading their docs it might be possible, but it would need to be setup with two separate daemons on separate ports (or two virtual IPs) and with two separate config files. It does have options to select the GPU to capture and monitor on the GPU (Sunshine/Moonlight are limited to 1 display due to the nvidia protocol), so it seems like it should be able to work. I have not tested it though. It's of course limited to 1 session per GPU and Linux only, but Linux gaming has come pretty far.
Thank you very much for this tutorial, apalrd : clear, simple and efficient. I set up the two desktops configuration I needed at home (Ubuntu 22.04 + Intel motherboard + Amd Radeon RX6600) and it worked like a charm (Wayland and/or Xorg). My only trouble is with seat1 : I can't leave or change a user session without totally crashing it + starting over a user session on seat0. Not very handy as a desktop configuration. Is it the same for you and, if so, do you have a clue why ?
3:40..
Is it MoBo "ASRock B450M Pro4"?
Does the motherboard support IOMMU?
9:50..
With this iGPU, does the following file exist?
File path:
ls "/sys/bus/pci/devices/0000:07:00.0/rom"
I ask because I am looking for hardware for a low cost homelab with headless hypervisor with single GPU passthrough.
Yeah it's the B450M Pro4. IOMMU setting is in Advanced->AMD CBS->NBIO Common Options->NB Configuration and defaults to disabled (of course) but once that's fixed it does support IOMMU in Linux.
I wouldn't say I picked it because it was a good choice, it was low cost at the time and fit my needs. I was using the iGPU for transcode for my security system at the time, but that was in an LXC container so I wasn't passing it through.
I've successfully passed through a GPU in the primary PCIe slot in Proxmox previously on this hardware. At that point I was using the Vega GPU as the BIOS GPU and also for Proxmox, so the second GPU was uninitialized. Looking at the IOMMU groups, it looks like at best the Vega GPU will only pass through without HDMI audio (since the USB root, PSP, and HD Audio all share a group with the HDMI audio controller, but the GPU itself is in its own group).
9:12
Ah yeah, there goes HDR haha
Wayland has been taking its sweet sweet time reaching feature parity (it's getting there!) but now with broader drm_lease compatibility (gnome merged protocol support just recently) I feel like that'd be ideal for sharing a single video card via logind multiseat? Possibly... I haven't given it a try - I'm mostly spitballing after seeing this video. I'm sure someone out there knows why that won't work. Like maybe compositors have yet to do some funny mode setting shenanigans? I'd love to know either way.
Do you know if it is possible to configure each HDMI, DP port as a seat? just one graphic card, several seats
No, since the entire GPU device is given to the X server, individual ports aren't treated as separate devices.
I kinda skipped through the video at this point, but I will save this and attempt to recreate it soon. Thank you!
Have fun!
@@apalrdsadventures Thank you. I was just hoping that this would be possible on a single GPU. I have done the weird GPU-PV thing that some youtubers tried a while back, but that has been abandoned and the new implementation by Microsoft just ignores AMD GPUs...
p.s.: Found your blog! VERY HELPFUL!
another completely unique topic
If open-source GPU drivers from AMD would allow resource sharing between multiple users then this solution would be interesting.
Of course I'm skipping GPUs from NV because in this case it's unrealistic.
(closed drivers)
There's very slight evidence that AMD is currently working on this, at least a way for multiple users to submit work to the GPU. I assume it's for shared compute / OpenCL reasons, but there is kernel work going on to amdgpu to deal with multiple users.
NV does their own whole mess on Linux
Linux Direct Rendering Infrastructure allows it. In user-space for example drm lease wayland extension exists.
Very nice tutorial thank you so much :-) For now I think I will stick with my pci passthrough but when I get into games that require non vm's I may try this solution.
PCI Passthrough is certainly a valid solution too
I love the title, 2 girls 1 cup
Just discovered this channel, great stuff.
Glad you like it!
i use aster multiseat and sandboxie plus application copier, works very nice with Minecraft once u get use to the setup wich adds 5 or so mins with turing stuff on and all that
You don't even need sandboxeie anymore
Thanks for the video, I needed it not for gaming but because my wife's computer broke and we didn't want to buy another. Ubuntu wouldn't work out of the box and I'm not good with computers, but Rocky Linux worked right away
Interesting! Just to mention, u can do gpu passthrough with qemu/kvm without wasting one gpu for the host. I do this with proxmox all time. Have 2 gpus plus igpu, run linux vm on igpu for proxmox webpanel, 2 additional vms on the gpus.
In my case I'm gaming on the iGPU as well, one seat gets the iGPU (so its 1 iGPU + 1 GPU). The Vega iGPUs are actually quite decent.
How do i attach ‘virtual devices’? I have 2 xbox wireless controllers connected via bluetooth. Currently both controllers is attached to seat0
I don't have any Bluetooth hardware to try this with, unfortunately. Do they show up as separate devices in seat-status?
Worst case, it's certainly possible to split them up if they are on USB or use their own USB dongles instead of Bluetooth.
I can't POST once I set IGD to be primary if I have a dedicated card connected to the motherboard. MSI Z690 motherboard running 13600K
Great content!
i've got multiseat working but i want to share it via the internet, there's this app called sunshine but i'm not able to assign the virtual devices that it creates to the 2nd seat
lol..2 gamers 1 pc... i wish there was a way to use a single GPU and a single CPU for multi-seat :(
Bummer. No amd and nvidia mix option !?
I really need one monitor multi seating. I tried Hyper-v with gpu partitioning("ruclips.net/video/XLLcc29EZ_8/видео.html") and worked but I cant get any way to multi seat with it. Tried every 3rd party pay option for mouse. Nun work as I cant see the curser. It's there just cant see it. :(
Any suggestions ?
Blame nvidia for messing with the gl / vk libraries instead of using the proper Linux display system, as to why multi seat doesn't work.
What are you trying to do with one monitor? Or do you mean one GPU?
@@apalrdsadventures I'm trying to get 3 gaming instances on 2 monitors. but cant get moues/keyboards to work to its own desktop. Vmwear doesn't have the correct direct-x :( But it does allow me to have multi mouse and keyboard to each vm.
FYI Wayland itself provides API for multi-seat (a.k.a. multiple sets of monitor+mouse+keyboard at once)
It's the same SystemD API that I'm using here, Wayland should correctly implement it but I found it didn't work correctly for OpenGL and especially Vulkan.
@@apalrdsadventures Technically Vulkan can do that with offscreen framebuffer, but the only problem is how programmers understand it (personally did I give up to code it last project). I don't know about OpenGL since Wayland only uses EGL for every OpenGL operations
With Wayland I had Vulkan rendering to the default GPU (the Raven Ridge) off screen and it was around half the speed the GPU should have been capable of (presumably due to the system memory copy of the framebuffer from one GPU to another), even without a game running on the Raven GPU.
With X the games were correctly running on the GPU for their seat. I'm assuming X is setting the mesa vk preferred GPU environment variable and Wayland is not, or it's not setting it differently per seat.
Good stuff
Would this still work as wayland becomes the norm on Linux?
Cool video, never heard of this.
I must ask! What keyboard you use???
The little keyboard+touchpad is a GearHead Smart Touch II, but it's not sold any more.
The nicer ones are a pair of Razer keyboard+mouse (they are my brothers), and I use a Unicomp IBM replica and Logitech MX Verticla mouse.
How taxing is this on the processor? As some games tend to be more CPU focus rather than GPU.
Very game dependent. Some games also tend to be CPU bound because they are single-threaded or lightly multi-threaded, so having 8 CPU threads in this setup can deal with 2 CPU bound games which only run on 2-3 threads each.
You can think of it like having two games open at the same time. It hugely depends on the games how taxing they are.
GREAT VIDEO!!!! THANKS FOR THE EXPLANATION!!
Glad you liked it!
Very Nice!
My friends and I played 4 players in the game aliens: colonial marines on one RTX2060 Super 8Gb video card. Windows 10 + ASTER will be cheaper for 4 players.
Thanks a lot for your help to us!!!
Using my 4090 on Windows 11 between 2 setups. With aster that's all u need
It's cheaper to get a pci-e x1 to x16 adapter and a 9600 gt than the cheapest license of aster (which is only for 2 seats). Which is good enough for games like xonotic, 0ad, Warzone2100, and maybe even Portal. If you want to be able to play something more resource intensive, yes, aster is the way to go.
I couldn't find a crack that works for aster, if you know one let me know.
I don't speak a lot of english, sorry if my comment has lots of grammar errors and stuff.
Does seat configuration are permanent?
this is how to maximize the use of a gaming setup
Linux numbah one
I got a headache trying to setup a multi-sest between amd r9 380 and nvidia 8600gts and couldn't make it go past the ubuntu shows on all displays and had only multiseat in text... so I bought aster 3 seats and am stuck with a windows 7 multiseat that works, only issue is gpu instability between amd/nvidia resulting in random reboots but disabling hardware acceleration in firefox about:config resolves that
My setup is 1 seat with amd r9 380, I game on that one, and 2 seats on nvidia 8600gts, one on svideo 720x480 32" panasonic crt tv, running kodi, the other a 40" hdtv in 1080p running kodi and/or firefox.
System can game on amd r9 and do 2 tv entertainment systems, i can run old emulators inside kodi that is fun and i have no problems running 2 kodi sessions on that old generation nvidia gpu. All Firefox gpu acceleration is disabled.
Thanks for your video. It is really amazing.
Glad you liked it!
A dream coming true ❤
Do i have to use different steam account for each user?
Sorta. I was able to use one account to play singleplayer, but it wouldn't sync my save games to both at the same time. So one seat would end up starting from the beginning. You also can't play multiplayer with your own account, so no LAN parties or co-op if you don't have separate accounts.
I am so excited.
I am too
@@apalrdsadventures I was just able to finish this video. I am very grateful you made it. I have kids and tried this myself for homework stations, but couldn't get sound to work at the second station. Or it was first come first serve with audio. A very frustrating experience. I'm going to try this again soon using your video as a guide.
Thank you again!!
Glad you enjoyed it! My next goal for this is to use DisplayLink adapters (USB3 laptop docks with USB3 graphics devices) which should bring the cost down a lot if you don't need 3D acceleration
very cool like always!!
Glad you enjoyed it!
@@apalrdsadventures You're insane like me :P in IT of course. Gload to see that passion :)
Why can there be a GUI for setting up multiseat?
Probably because it's not a very common setup for users
i wish there was, it was a pain for me to try multiseats and didn't work for me, i had to purchase a licence of aster for windows 7 which is now not supported anymore, i wish it would work on linux the easy way too
Yes. Too bad taht whenever we start to get all the kinks ironed from what is used (xorg) they have to be switched up because of security (wayland) forcing everything to lose 10 years of development every switch.
20 years of evolution, 10 years of devolution, always like that.
really nice
Is it possible to split a single monitor in half in this way?
Not easily, this method splits across driver instances, so it needs a separate GPU and separate outputs per seat
xD why is this so interesting Thank you very much for the content
Glad you enjoyed it!
Bro i am begging you please make a video on how to do this on windows 11 home without virtualization and without aster multiseat
super cool
Thanks!
Is it possible to do more than 2 players?
Yes, with one GPU per player
I love this channel
Thanks for the support!
is there a way of using hypervisor, and assign a vm to a seat?
Not directly
If you're using a hypervisor, you would pass through the entire PCIe device using vfio instead of a method like this. The GPU would be bound to the VM, and the VM would control it completely
@@apalrdsadventures in that case, i guess multiseat wouldn't be necessary?
Not if you use PCIe passthrough - at least not using this method. You'd still have a multi-seat system, but via virtual machines which partitioned access to the hardware.
Nice
Thats kinda hard lol, I just made a qemu vm with gpu passthrough, not as efficient but much more easy
very cool. but shame not many games run on linux
I hope wayland gets better multiseat support soon.
It already kinda works, but gdm gets frozen when logging in on the second seat. Also, I think gnome-shell refuses to run in second wayland session, but you can still run another compositor like weston or even kde plasma
I wish I could do this exact thing on windows :(
It's more of a 'Microsoft doesn't want to let you' than 'it's technically impossible' with Windows, since it absolutely supports multiple user sessions over RDP (with a server license of course).
you can if you install aster and maybe buy a 2 or 3 seat licence which costs less than $100. I run it with a amd r9 380 and an old nvidia 8800 gts for a retro pc with svideo tv out
Pls keep making vids
good vid
thx
In virtualization the host don't need a GPU or APU 🙂
POV "Abriendo el Google Chrome"
You should have started MUCH simpler. Just one graphics card - maybe even one of those Intel ones. Either NO acceleration for BOTH seats or acceleration for seat0 and no acceleration for seat1.
Would this be possible? Much easier? How?
It's much easier to split a system up by GPUs than to split a single GPU into multiple seats, mostly due to the architecture of the Linux DRM/DRI being card-based and not port-based.
Each card needs to be 'owned' by an X server, which runs in the user session. Any app which wants to perform accelerated rendering needs to go through the X server for authorization to access the card's render functions. For two user sessions to share a card, you need to spawn a third global session and forward the rendered output of each user session, and the user sessions wouldn't get acceleration access. This is a much more complicated setup than just having two cards = two X servers = two users.
Hey guys .
I've some suggestions .
Why don't you guys try WSL ( Windows subsystem linux )version 2 .
I don't assume you guys use Windows , but this is too awesome to pass up .
It lets you install a linux distribution , with way too much cooperation with windows , such that you can drag linux windows through windows !
it shares the hypervisor of the windows , and doesn't stack or nest hypervisors or kernels , but the linux and windows kernel works alongside each other .
I want to see if you can multiseat using a windows software but through that , run linux software with just one gpu .
it's not related to the discussion here, you're talking of trying linux on windows.
@@joelpichette its a little different than trying linux on windows .
In some ways better even .
It kinda works like wine ( acronym for ' wine is not an emulator ) .
Helps run windows software on linux .
i dont have iGPU
You can use two physical GPUs too, or more
Two GPU ? GPU more expensive than CPU....none sence for this information.....
I'm saving money on the rest of the system by only having one system instead of two of everything (CPU, RAM, power supply, motherboard, case, ...)
Brother is it free ??
The software is. You need two GPUs still though.
Can you still do it with one GPU?
Not using this method