Boot to virtual PC from a terminal (with only monitor, keyboard and mouse)

[Chiyawa]

Hi,

 

I'm new to virtual PC, so I need some information.

 

I'm planning to build a server that host virtual PC for my home. Planning to use 2 virtual PC currently. I'm planning using a normal PC (Ryzen 7, 32GB RAM, 2 Graphic card) to power the system. So my question is how to do it? The terminal only have a monitor, mouse and a keyboard, connect to a extender hub that runs ethernet cable to the PC which terminates with VGA and a USB.

 

Thanks in advance.

[Windows7ge]

Going to need a little more information. So you want to build a Virtual Machine Server and have it host a couple of VMs that I presume will be used for gaming or other things? First thing you're going to need is a hypervisor (the OS). Do you have one in mind currently?

[Chiyawa]

3 hours ago, Windows7ge said:

Going to need a little more information. So you want to build a Virtual Machine Server and have it host a couple of VMs that I presume will be used for gaming or other things? First thing you're going to need is a hypervisor (the OS). Do you have one in mind currently?

I don't have any specific hypervisor OS. I'm thinking of using Linux, but am afraid of drivers incompatibility (not very familiar with Linux). I think I'll be sticking with Windows 10 Pro and then put Windows 10 Home as the virtual PC. I tried to use a video from LTT as the guide, but this video only shows the wiring stuff and they login directly to the PC.

 

Also, I'm still gathering data, so I don't have a good knowledge on the system, but I have made up my mind to use Ryzen 7 3700X or 3800X and 32GB of RAM. Each Virtual PC will get 6 threads processor and 8 GB of RAM. Also, this system will host as simple network storage.

 

Sorry, this is my first time attempt to do such a crazy stuff.

[Windows7ge]

Are you looking to give the VMs GPUs or are these VMs going to be for purely storage & other that don't require GPU acceleration?

[Slayerking92]

https://en.wikipedia.org/wiki/Thin_client

 

[Chiyawa]

10 hours ago, Windows7ge said:

Are you looking to give the VMs GPUs or are these VMs going to be for purely storage & other that don't require GPU acceleration?

Uh, the host will be the storage server as well as the host for virtual PC, so I can delegate the GPU to the Virtual PC.

[Chiyawa]

9 hours ago, Slayerking92 said:

https://en.wikipedia.org/wiki/Thin_client

 

Oh, I'm not looking for thin client for this setup. I will be directly connect the display, mouse and keyboard to the PC, you know, like creating multiple PC from one real physical PC.

[Windows7ge]

3 minutes ago, Chiyawa said:

Uh, the host will be the storage server as well as the host for virtual PC, so I can delegate the GPU to the Virtual PC.

What GPU are you planning on passing though to the VM?

[Chiyawa]

I don't have a clear cut, but maybe RX580 or 590.

[Chiyawa]

5 minutes ago, Windows7ge said:

What GPU are you planning on passing though to the VM?

I don't have a clear cut, but maybe RX580 or 590.

[Windows7ge]

4 minutes ago, Chiyawa said:

I don't have a clear cut, but maybe RX580 or 590.

I was going to say if you said NVIDIA you were going to be very SOL. NVIDIA installed a "feature" on their desktop series cards where when you go to install the driver it detects if it's running in a Virtual Machine. If it does it disables itself and reports Error 43 in Device Manager. You'd have to either make the card think it's not in a VM (a hack with mixed success) or buy a Quadro/similar series card.

 

As for AMD. They don't care. Some GPU's pass-though and work better than others. For me the R9 290X works pretty well in a VM. Depending on how much the 580/590s differ it may work fine or not.

 

As for the hypervisor I only know of one for Windows that allows for GPU pass-though and that's Hyper-V. Unfortunately I only know of it existing on Windows Server. A little Googling says it can be installed on Windows 10 Pro though.

 

If you go this route though I won't be able to help much. I've never done GPU pass-though using Hyper-V.

[Chiyawa]

1 hour ago, Windows7ge said:

I was going to say if you said NVIDIA you were going to be very SOL. NVIDIA installed a "feature" on their desktop series cards where when you go to install the driver it detects if it's running in a Virtual Machine. If it does it disables itself and reports Error 43 in Device Manager. You'd have to either make the card think it's not in a VM (a hack with mixed success) or buy a Quadro/similar series card.

 

As for AMD. They don't care. Some GPU's pass-though and work better than others. For me the R9 290X works pretty well in a VM. Depending on how much the 580/590s differ it may work fine or not.

 

As for the hypervisor I only know of one for Windows that allows for GPU pass-though and that's Hyper-V. Unfortunately I only know of it existing on Windows Server. A little Googling says it can be installed on Windows 10 Pro though.

 

If you go this route though I won't be able to help much. I've never done GPU pass-though using Hyper-V.

I see. Thank you.

 

Yeah, I did some research before hand, so while I have a little knowledge about this setup, I'm still looking for ideas and more information. Thanks for helping me.

[Windows7ge]

1 hour ago, Chiyawa said:

I see. Thank you.

 

Yeah, I did some research before hand, so while I have a little knowledge about this setup, I'm still looking for ideas and more information. Thanks for helping me.

What motherboard are you going to use? To pass-though a device to a VM you need to enable IOMMU groups. This is a function/feature that I know Threadripper has but I don't know about the desktop series platform.

 

Is the host going to be used like a daily driver or will its sole use be for managing the VMs?

[Chiyawa]

2 hours ago, Windows7ge said:

What motherboard are you going to use? To pass-though a device to a VM you need to enable IOMMU groups. This is a function/feature that I know Threadripper has but I don't know about the desktop series platform.

 

Is the host going to be used like a daily driver or will its sole use be for managing the VMs?

As for IOMMU, I think most X570 motherboard can be used. I'm currently using Gigabyte X570 Aorus Pro WiFi, and the motherboard seems to have this feature.

 

The host will be a daily driver, as it'll be managing storage as well.

[Windows7ge]

6 hours ago, Chiyawa said:

As for IOMMU, I think most X570 motherboard can be used. I'm currently using Gigabyte X570 Aorus Pro WiFi, and the motherboard seems to have this feature.

 

The host will be a daily driver, as it'll be managing storage as well.

Then you seem to have most of what you need figured out. Anything you're currently not sure about?

[phoenixflower]

On 1/6/2020 at 8:44 PM, Chiyawa said:

Hi,

 

I'm new to virtual PC, so I need some information.

 

I'm planning to build a server that host virtual PC for my home. Planning to use 2 virtual PC currently. I'm planning using a normal PC (Ryzen 7, 32GB RAM, 2 Graphic card) to power the system. So my question is how to do it? The terminal only have a monitor, mouse and a keyboard, connect to a extender hub that runs ethernet cable to the PC which terminates with VGA and a USB.

 

Thanks in advance.

If you're willing to put in some leg work then using something like Proxmox will end up being a great utility for you.  It's just Debian with a lot of IT virtualization knowledge packaged with it.  It has a lot of sane defaults and also supports ZFS if you want it to manage multiple drives.  If you want to virtualize FreeNAS (give it minimal 8 GB of RAM) to manage ZFS in a safer way then that could also be an option.  You should probably daily drive Proxmox as your storage solution for a while to make sure it's what you want.  Sometimes ZOL (ZFS on Linux) can have very unfun moments.


If you just want something that's going to work OOB and willing to give up stuff like ZFS and iSCSI then UnRAID is usually pretty acceptable.  The only quality of life thing(s) they haven't done in my humble opinion is allowing their implementation of the Kernel to actively allow the kernel to give up and re-take PCI-E hardware.  If you go online a lot of very reputable people (including SpaceInvader) will simply recommend to blacklist the parts you want to passthrough and although that does work.  It does come with consequences.  LIke when everything goes wrong, I like knowing I can restart my machine and end up back at the command line of my host from one of the graphical solutions.  Proxmox does in fact support that headless configuration in which you can use the hardware at boot but simply give it up to VMs over time.  They also pick up on stuff like NAVI reset patches pretty fast.  Some members of TPP (The Passthrough Post) came up with a temporary patch to the AMD reset bug for their GPUs and it appears to be within Proxmox.

also from my experiences, running docker in LXC is less of a pain then running docker natively because it always tries to share the host's ip and fights over port 80 and port 443.  UnRAID has it's upsides but the downsides are all pretty comical.  They're all "WHY" problems.


If you want audio to work in your VM you will have to do one of three things.  Rely on the Audio out from your graphics card, buy a USB Add in Card.  If you would like a recommendation I can certainly do that or finally use an Airplay Server to pass stuff from your VM to your Linux Host.  The latency isn't terrible because the traffic shouldn't actually leave your Network Interface Card.  There's a lot of Pulseaudio tutorials on how to do audio networking, Pulseaudio is notorious for having worlds of latency even in the best of times.

They're a variety of ways of sharing your mouse and keyboard with your host and VM.  The easiest is just to use the web interface.


If you decide to use Proxmox I can provide an example config file that will hopefully help with some of the nitty gritty stuff.
If you choose UnRAID then I can provide context clues and SpaceInvader tutorials

I personally run the x570 taichi and it's a virtualization dream boat, you will have to cool it  correctly if you go that route because if you're not careful in Virtual Function Input Output (VFIO) territory the chipset can get exceedingly hot.  The x470 boards are also valid candidates.  I have only had reasonable success with the AsRock deployments but it's been a while and I'm set to test a gigabyte board later this week. Fingers crossed*

 

It's a lot to digest.  Feel free to reach out if you have any comments, concerns, opinions

[Chiyawa]

4 hours ago, phoenixflower said:

If you're willing to put in some leg work then using something like Proxmox will end up being a great utility for you.  It's just Debian with a lot of IT virtualization knowledge packaged with it.  It has a lot of sane defaults and also supports ZFS if you want it to manage multiple drives.  If you want to virtualize FreeNAS (give it minimal 8 GB of RAM) to manage ZFS in a safer way then that could also be an option.  You should probably daily drive Proxmox as your storage solution for a while to make sure it's what you want.  Sometimes ZOL (ZFS on Linux) can have very unfun moments.


If you just want something that's going to work OOB and willing to give up stuff like ZFS and iSCSI then UnRAID is usually pretty acceptable.  The only quality of life thing(s) they haven't done in my humble opinion is allowing their implementation of the Kernel to actively allow the kernel to give up and re-take PCI-E hardware.  If you go online a lot of very reputable people (including SpaceInvader) will simply recommend to blacklist the parts you want to passthrough and although that does work.  It does come with consequences.  LIke when everything goes wrong, I like knowing I can restart my machine and end up back at the command line of my host from one of the graphical solutions.  Proxmox does in fact support that headless configuration in which you can use the hardware at boot but simply give it up to VMs over time.  They also pick up on stuff like NAVI reset patches pretty fast.  Some members of TPP (The Passthrough Post) came up with a temporary patch to the AMD reset bug for their GPUs and it appears to be within Proxmox.

also from my experiences, running docker in LXC is less of a pain then running docker natively because it always tries to share the host's ip and fights over port 80 and port 443.  UnRAID has it's upsides but the downsides are all pretty comical.  They're all "WHY" problems.


If you want audio to work in your VM you will have to do one of three things.  Rely on the Audio out from your graphics card, buy a USB Add in Card.  If you would like a recommendation I can certainly do that or finally use an Airplay Server to pass stuff from your VM to your Linux Host.  The latency isn't terrible because the traffic shouldn't actually leave your Network Interface Card.  There's a lot of Pulseaudio tutorials on how to do audio networking, Pulseaudio is notorious for having worlds of latency even in the best of times.

They're a variety of ways of sharing your mouse and keyboard with your host and VM.  The easiest is just to use the web interface.


If you decide to use Proxmox I can provide an example config file that will hopefully help with some of the nitty gritty stuff.
If you choose UnRAID then I can provide context clues and SpaceInvader tutorials

I personally run the x570 taichi and it's a virtualization dream boat, you will have to cool it  correctly if you go that route because if you're not careful in Virtual Function Input Output (VFIO) territory the chipset can get exceedingly hot.  The x470 boards are also valid candidates.  I have only had reasonable success with the AsRock deployments but it's been a while and I'm set to test a gigabyte board later this week. Fingers crossed*

 

It's a lot to digest.  Feel free to reach out if you have any comments, concerns, opinions

Wow, that's a long one. Thanks for the info.

 

Anyway, this is just a little project that I intend to build, but I can't believe the learning curve is quite steep. Well, you are learning everyday.

 

Yeah, I haven't thought about the audio. maybe I need to dig into that, but probably later.