November 7, 20241 yr I currently have an Unraid server at home that runs a bunch of docker containers for various home use things like a nextcloud server, jellyfin, audiobooks, game servers. I work at a veterinary clinic and we have a windows server that runs our PIMS software and some network shares. We would like to put computers in each of our exam rooms. The problem comes because we have a CT machine that takes pretty significant graphical and CPU power in order to run the reconstructions in order to view the scans. We could just buy a high end desktop for each room, but that seems like it would be fairly expensive and not very expandable. The other option that we have discussed is purchasing another server (or building it) and running one or multiple VM's to be able to allocate resources and just run a very small bare-bones computer in each exam room and connect via RDP to a VM on the host machine. I have 2 major options from here, run a single VM that is dedicated only to viewing CT scans, but can only be used in one room at a time, or running a VM for each room and running very bare OS for RDP and doing all computing on the VM. From what I can find however it looks like on unraid GPU's can't be shared between VM's running at the same time. I have considered doing something like VMware, or another hypervisor, but there are other docker containers that I think would be useful to be run on the same server. Any thoughts?
November 12, 20241 yr Community Expert you can definitely set up a VM with terminal access as a connect monitor to unraid server and use it as a terminal host or guest access via rdp systems like parsec. review this: https://rigaku.com/products/imaging-ndt/x-ray-ct/learning/blog/consider-buying-computer-ct-image-processing I'm quite aware of some caveats and issues when dealing with medical HIPAA stuff and hardware you may face with the CT imager... to accomplish this, you will need a Graphics card in the unraid server. Something like a cheap ebay purchase of a 20 series card like the nvidia 2060 or nvidia 3060 would be greater here. As I actual had a similar instance recently setup... It was actually for this reason (as I work at a university for a lab) that I went to Proxmox and VGPU by splitting a card like the 2080 or 3060 into multiple VM. Then virtualized unraid with a HBA. Just throwing this option out there to you. it's also not hard to move unraid working samba/dockers to a different system... (I still hold unraid as docker king in terms of easy to implement and use any docker system...) In regard to your use case, yes, Unriad is a hyper v and can act as a vm with gpu passthorugh. BUT what your describing is more vgpu related task where one can use parts of the g card in different ways when the demand is there... You are correct, in unraid 1 graphics card to 1 vm. with vgpu we can split the G-card based on its vram. Example A 12GB nvidia 3060 can be split multiple ways into example 4 different acting g cards with 4 GB of video ram...(if splitting this way the vram vgpu can't be mixed so if you go 2x 6 gb vram you may get 2 but 1 for sure. Lower the vgpu instance the more with less vram. With CT imaging its recommend to be min at 4GB ideal 8GB...) so with Unraid, it is 1 vm with 1 gcard passthorugh via VFIO/bind, BUT one could use a multi-user RDP program such as parsec ... Parsec can allow multiple user to connet to the VM and use the machine BUT all would need to share the resource and use it 1 at a time... Sadly, I'm not aware of any vgpu support in unraid at this time. mby in the future with how open source driver just dropped for NVIDIA vgpu. While one could purchase a vgpu card such as the Tesla t4...https://www.amazon.com/PNY-Datacenter-Express-Passive-Cooling/dp/B07QC7V1KS ^ Example Unraid has no way that I know of to get the nvidia vgpu stuff installed. also with Nvidia stuff one would need to go through there license portal which is a pain. mainly for vgpu host driver and guest drive software installs. AMD is similar... ############################### But it really sounds like you need a different system for VM access and terminal support with resources... Unraid is nice and all but to me it more catered towards home lab users not enterprise level uses cases such as 45 drives cockpit or truenas scale is... it sounds like you may need a new workstation system if you want to keep unraid samba /docker working on curent... So lets review A Quick build as sadly price on hardware form 3rd party... will be anything between $1000-2000 range... So here are some Examples: amd epyc gen1 rome.. https://www.ebay.com/itm/134674583565 with tesla p4 (pascal) (nvdia just killed vgpu pascal so alot of theses cards will be coming up if your ok with working with EOL hardware and using workarounds.... this is the build i just finshed making with proxmox vgpu on a tesl p4 on epyc...) -- Just showing that it is still possible... so g-card vram is everything here...(as it is the main device being used to help process the ct images... tesla p4: https://www.ebay.com/sch/i.html?_from=R40&_trksid=p2332490.m570.l1313&_nkw=tesla+p4&_sacat=0 or 3060: https://a.co/d/6cKKvow *this is not including other extras that may be needed such as the need for a PC case, power supply, Storage HD etc... But that building new by the time you get xyz part it may be worthwhile to sink into a prebuilt: Example 45 drives cockpit... https://store.45homelab.com/configure/workstation Example of top of line workstation... https://bizon-tech.com/bizon-v3000.html?srsltid=AfmBOoqqtDA3RSyNJYhmZPdrHU9JMLeJstsacKhL04XT5Q6VNlvCW8qxm3U But, I would personally point you towards proxmox using vgpu and setup your multiple VM as needed. Edited November 12, 20241 yr by bmartino1
November 12, 20241 yr Community Expert If you go proxmox vgpu, you can have your employee/clients with laptop or other machine taht can connect to the multiple VMs that the machine would allow for... the server can then be somewhere tucked away for power and network only and you would use tools like parsec / Microsoft RDP / VNC to access the VM... a 3090 with 24 GB of Vram can easily be split into 3 VGPU of 8 GB systems per the example. FYI: With proxmox its either vgpu or g card passthorugh 1 or the other not both... you can easily virtualize truenas scale with a hba for docker / samba server. or run a proxmox lxc / VM for docker as a seperate instance... it comes down to what you want it to do and how you want to interact with it.
February 14, 20251 yr I am also considering using Unraid in business. The main thing is to choose the proper hardware, especially if you need a powerful GPU. I want to understand which cards work best for virtualization and transcoding. I often learn about new products in this area on the DDI blog, where there is much interesting information about server solutions. If anyone has experience with Unraid and GPUs, it would be helpful to hear some advice.
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.