sprange

I own a small dental practice. Since 2010 I have used ESXI for server virtualization although my demands have been extremely limited--mysql and various programs that require USB security dongles. Over the years I've had good quality server hardware randomly die and had very little downtime (thanks to VMWare Workstation on a spare computer back when I had only one host). The trend in dental software is cloud based solutions in which your patient's data is basically a service and access to records is not guaranteed. Where I practice I need to have records available for at least 30 years so local storage and programs that may not be supported forever but at least can run in a VM are a must. I have several workstations that could easily be backup hosts for ESXI however the workstation software (dental 3D modelling for example) require GPU and USB passthrough, both of which are hit/miss with ESXI. I have a Zenith Extreme 2950X computer for example that will not work for this purpose. Unraid however has no problems passing through the GPU and either a USB controller or more conveniently just a keyboard, mouse, and a dongle. For my purposes the VM speed is just as fast as baremetal. The cool thing is that I can install ESXI inside UNRAID (complete with hardware RAID controller passthough) and it just works. A VM running in ESXI under Unraid benchmarked faster than a VM under Proxmox baremetal which was surprising. Unraid isn't an ideal product for my needs as it is, but it sure opens up hardware that would otherwise be useless for ESXI. I understand that Unraid utilizes KVM. For kicks I've used Gentoo to create the most optimized host for VMs but it was extremely brittle and was way too much effort. With Unraid I could easily pass through GPUs and USB controllers FAST. Unraid sometimes gets called a gimmick etc but not only can it turn a 3970X into three gaming machines at my home, but also open up hardware at my office, Works for me!

Tossing <boot order='1'/> (or 2 if I have a bootable CD image) is a necessity for my VMs that have a nvme passed through to them. After making any changes to a VM via a form (which can be easier than XML editing for certain tasks) erases this line. Definitely would be appreciated if it could stay. Bonus points if you can add it without resorting to XML--would save tons of 'HELP' posts for users who also pass M.2 nvme drives through to VMs.

Agreed 100%. I don't need to reboot my server that often, but when tinkering with VMs it becomes a necessity. Waiting for this service drives me mental!

I have this motherboard coming since my Aorus Extreme died randomly. This motherboard has Intel chips X710-AT2 and i225. From what I can tell the i225 is apparently supported in the latest beta. Getting the X710 working would likely require either a new kernel or module or both. I actually couldn't find much about Linux support for this chip.

I am pretty sure I have used most of those settings before (not all) and I don't recall seeing any benefits although I never specifically measured FPS for example. One thing I do note is the topology...what's up with that? Certainly doesn't mirror the host CPU.

I have a s limilar setup (3970x with 3 2080TIs) and also a spare (2950x with 3070 and 2080TI) and they all run games at very respectable frame rates. The only real difference I see with your setup is the lack of passed through nvme (I have dedicated ones for each VM) and your clock setting is different. I use the default xml code (basically just has hpet no hypervclock yes). From what I've read clock settings can really influence Windows VM behavior. Maybe something to check? It seems mine are suboptimal but my results don't reflect that.

I tested mine out--works fine. Was able to sync to a wireless controller and play a game. The adapter is connected the USB adapter on my video card.

So you mean the new VM (created using the GPU passed through for the install and no VNC driver) still gives you a code 43 in windows? Do you have CSM enabled? I have nothing special in my config...just the vendor name to some random 12 digit ID and nothing else changed. Audio device passed through as well? Is the 1050 your only video card?

I generally only have good luck passing through GPUs when I create a new VM and do not use the VNC driver initially but instead the GPU (I have a 3970x with 3 2080TIs as gaming VMs). Also Q35 5.1. In my case I pass through M.2 SSDs but I've used virtual disks before with no problem.

I have the TRX40 Aorus Xtreme and have similar results. One of the controllers can be passed through fine. Another one results in the VM hanging at the boot screen for about five minutes before resuming. I haven't tried the Asmedia USB c port--I can always try that and use a hub.

I found the same when I used the USBc ports on my 20xx cards on my 3970x (I had usb hubs attached to them). Wasn't an issue with my 2nd generation threadripper though.

Ahh...I see you have <boot order ...> contained within the <source ... > tag. Rearrange it like I have mine and see if that helps...

Here is how my PCIE NVME drive looks in the XML code: <hostdev mode='subsystem' type='pci' managed='yes'> <driver name='vfio'/> <source> <address domain='0x0000' bus='0x02' slot='0x00' function='0x0'/> </source> <boot order='1'/> Yours is a SATA NVME so I have no experience with them. Maybe look through the forum as to how people pass SATA controllers but it absolutely is possible.

I wasn't sure what would happen in your case. If your nvme is connected via SATA then you should be able to boot it just like any other drive. The clover image was for cases where the NVME drive was passed to the VM.

The trick here is it edit the xml and find your nvme. Below source add <boot order='1'\> (just off my head might have the syntax wrong). This will allow the nvme to boot without always having to select a boot device. No modified clover image necessary. Regarding viewing your VM, you can always enable RDP and access it that way...that's how I do it. I take no credit for the boot order fix, I stumbled upon it in the forum after ripping my hair out. The catch is every time you switch to form view you will lose this line and you will have to edit the xml again and add it. Hopefully unraid will fix this eventually. I'm at work and don't have time to elaborate but if you are confused search for nvme and boot order and you'll find the posts I referenced this from. EDIT So I looked through your xml and details but it doesn't look like you are passing through a m.2 NVME like I thought you were. Perhaps it will still work for you. Find whatever device your NVME is (somewhere in your xml). Lets pretend it is this (likely not) <address domain='0x0000' bus='0x05' slot='0x00' function='0x0'/> place <boot order = '1'/> after that line for the appropriate device With this you will not need to use the modified clover...

I've used Unraid for a few months with my old x399 and 2950x prior to upgrading to a 3970x. Same 3x 2080 Ti setup. I had experimented with gentoo and arch as server hosts but neither offered any advantage other than access to newer versions of libvirt and qemu. My only need for Unraid is as an easy to deploy kvm host. To that end it is hard to beat, especially with community plugins. I also tried out multi seated setups for windows (aster) and linux but I was surprised that two virtual machines seem to be much more responsive than a multi seat configuration. No benchmarks to back that up however. The only thing I'd like to see is striped storage for my VMs. I noticed the network psuedo connection is 100Gb so host virtual network operations won't be arbitrarily restrained, like when I used kvm on gentoo. Qemu 5.0 also has direct access to host folders which might be easier than NFS. Anyway I really like the implementation of Unraid. Good work!!

The TR and X399 generally work quite well for gaming (I currently have a Zenith Extreme and a 2950x). GPU passthrough is easy enough, but there -may- be reset issues with your AMD cards. I think I came across a post talking about using a custom kernel with a patch that addressed this problem. You don't necessarily need another video card as once you have everything set up you can stub the video card in slot 1 and your server is still functional although you'll need to connect to it through another computer. With another video card chances are one of your 5500XT cards will be in a x8 slot which technically doesn't matter. No recent Asus x399 motherboard allows you to pick the primary GPU slot so it defaults to slot 1 and there goes your main x16 slot. With my 2950x I can run three gaming VMs extremely well (two connected to monitors and one is used for streaming but is still connected to one of the other monitors). Sean

Do you need a video card for unraid? I too have a X399 motherboard and utilize slot 1 for a 2080TI (it has a waterblock so only occupies two slots instead of 3) that is passed through to a VM. I just put the device id in the kernel line for VFIO so the device is stubbed as soon as the kernel loads--no need to blacklist etc. For the first slot I do however have to provide a bios image for the card, the ones in other slots do not require it. Sean

On my threadripper 2950X (being replaced by a 3970X picked up on Craigslist for what must have been something as drastic as a divorce) the two 'halves' of the CPU (so cores 0-15, and 16-31) communicate directly with different PCI-e slots and devices. So for best performance it's best to pair a passed-through device with a VM utilizing the cores with the direct connection. This varies with CPU types however.

I'm able to passthrough two 2080TIs, and also 3 M.2 NVME drives. The only tricky part was determining their order to match with the nodes on my AMD cpu. My machine has no primary video card as well.

Thanks to SpaceinvaderOne's very handy video on NVME passthrough to a VM guest and with the modified Clover image I was able to boot W10. However it always has the wrong boot order so I have to manually boot the NVME drive. Changing the boot order doesn't stick through boots. Any suggestions as to how to make it boot the NVME without user intervention? I may revert back to VirtioSCSI as the loss in performance is made up for the convenience of being able to easily relocate drives. Anyway I keep coming back to Unraid because it handles VMs very well (I have three VMs with 2080TIs passed through) and setup and remote access is a breeze. Sean