Daniel15

I wasn't aware of this chipset. Thanks for the info!

I have the MicroATX version, which has the embedded IPMI. It works well for me. It's not the right choice for everyone (it's only got one PCIe 16x slot for example), but it was a good fit for me since I needed a MicroATX motherboard for my case (Fractal Design Node 804) and the only add-on cards I've got are a 10Gbps network adapter and a Coral TPU. For what it's worth, a huge number of game hosts are using this board. This board and a Supermicro one are the two most common boards with the W680 chipset. The issues with Intel 13th and 14th gen CPUs were uncovered by game hosts running W680 boards discovering that a huge number of their servers weren't stable. The Core i5-13500 is very stable. It was built on the 12th gen architecture (it's essentially just a 12th gen i5 with efficient cores added) so didn't have any of the issues the other 13th and 14th gen CPUs had. I don't think the 14th gen refresh had any of the issues either? It's unfortunate that there's no newer workstation-grade chipsets that support ECC. W890 only supports Xeon CPUs, and the consumer version (Z890) doesn't support ECC.

You can do this in the UI instead of modifying the command line arguments. Instead of selecting "None" for the network type, select "Container". It'll show a dropdown list to select the container to use.

IOMMU is what allows PCIe passthrough. The IOMMU section in Unraid allows you to bind an IOMMU group to one virtual function (VF), to pass it through to a VM. SR-IOV is what allows one physical device to be split into multiple VFs instead of just one. This is so it can be passed through to multiple VMs, or used on both the host system and one or more VMs. The most common use cases are for GPUs (like with this plugin) and for Ethernet adapters.

No - the native i915 driver doesn't support SR-IOV. The xe one does, but only for certain devices.

It does now - it includes patched versions of both i915 and xe. xe support was added in https://github.com/strongtz/i915-sriov-dkms/pull/344 I saw the disclaimer but AFAIK this is the repo the Unraid SR-IOV plugin uses.

Does it not use the xe driver from this repo? https://github.com/strongtz/i915-sriov-dkms

Is there a way to ensure all my current Docker containers are assigned their current MAC addresses before performing the update?

For Docker Compose, the existing plugin is pretty good. It's definitely not perfect (for example, it doesn't support override files), but the integration into Unraid's Docker UI and the folder plugin is pretty good. I create one folder per compose stack to keep the containers organized. I'd love to see a native inplementetion that's very similar. I'd also love to see LXC as a native feature, rather than having to use a plugin. For backups, I'd love to see a leading backup solution (either Borgbackup or restic) implemented natively, with a web-based config UI. The Borgmatic Docker container is pretty good in the meantime though.

The drive was missing from the ZFS pool itself. zpool status only showed one drive. It immediately started resilvering after I added the drive back using zpool add. I verified against a backup, and the config hadn't changed.

I recently had a power outage. My Unraid server is connected to a UPS and properly shut down during the outage. When I started it up again, I got this error when trying to start: data is a ZFS mirror with 2 x 20TB Seagate Exos X20 drives. It turns out that for some reason, one of the drives was no longer in the pool! I had to re-add the missing drive to the pool, remove Unraid's config for the pool, then re-add it in the UI. Now it's working fine and the drive that was missing is resilvering. I left it overnight and now it's around 50% done. Unraid should have a better error message than just "invalid config". Why is the config invalid? In my case, I think it's because the state of the pool in Unraid's config (two drives) didn't match reality (just one drive). What could have caused this? I searched the forum and saw there's several monitoring scripts for ZFS (which is something Unraid should really have built-in!) but they only look for degraded or faulted pools, which wouldn't have caught this. The pool was in an online state, just with one drive completely missing.

Is it ECC or non-ECC? If it's ECC then you should be able to see any error correction/detection in the syslog. If you see a lot then the memory is probably faulty. In either case, try run Memtest overnight and see if it shows any errors. I wouldn't trust generic brands from eBay. It's possible they're using poor-quality RAM chips that failed QA and were rejected by another brand. What's the brand on the RAM chips themselves? Have you tried them in a different computer?

The newer GPUs will only support SR-IOV via the xe2 driver, not via the old i915 driver (which is what this plugin uses). i915 isn't really being updated any more. The new driver provides native support for SR-IOV out-of-the-box, but only for some GPUs at the moment. As of Linux 6.18, SR-IOV in xe2 is enabled for Tiger Lake (11th gen), Alder Lake (12th gen), Raptor Lake (13th and 14th gen), and Arctic Sound (data center GPUs). However, I'm not sure if Unraid include this driver in their kernel build. There's a long thread about it here: https://github.com/intel/linux-intel-lts/issues/33

I've got a very similar combination in mine - can't remember the exact model numbers, but I have 2 x 32GB Kingston server ECC 4800MT/s and 2 x 32GB Kingston server ECC 5600MT/s for 128GB total. They're mismatched because when I upgraded from 64GB to 128GB a while back, the 5600MT/s RAM was the same price I paid for the 4800 a year earlier. It works totally fine, and runs at the lowest speed (so if you have both 4800 and 5600 RAM, it'll run at 4800). You may need to manually set the RAM speed in the BIOS if it's not automatically detected. The first boot with the mixed RAM may take a minute or two before Unraid starts, while it trains the RAM.

You can follow the docker-compose instructions as long as you install the "Docker Compose Manager" plugin (which will let you use docker compose). Be sure to change the paths in volume-settings.yml and lmd-settings.yml. There's a few containers so this would be non-trivial to set up in Unraid without using Docker Compose.

Use ipmitool (on Linux) to reset the password. https://www.intel.com/content/www/us/en/support/articles/000055688/server-products.html Does your USB drive work for booting on a different computer? It might be the USB drive that has issues.

It works from WinPE, so you can use something like Hiren's BootCD PE or Sergei Strelec's WinPE. Essentially a bootable Windows image with a bunch of useful tools preinstalled.

Yes. I'm using a Core i5-13500 in mine and ECC is active. All the Core i3/i5/i7/i9 CPUs support ECC. Intel artificially limit it to the workstation and server chipsets (like the W680) just for product differentiation - they need reasons for people to pay more for higher-end chipsets. AMD take a different approach and let motherboard manufacturers enable it for Ryzen CPUs, but only officially test/validate it for Epyc (server) and Threadripper (workstation) CPUs.

Also make sure you enable C-states, to reduce power usage. I'm using an NH-D15 in mine, in dual fan mode, along with some Kingston DDR5 ECC server RAM, and didn't have issues. I'm using the Micro ATX version, but RAM clearance should be the same for the full ATX version. Even tall RAM can work, but you'd need to remove one fan. Max RAM height is 32mm in dual fan mode and 66mm in single fan mode. The cooler was a tight fit in my case (a Fractal Design Node 804) with maybe 2mm of clearance between the cooler and the side panel, but it did fit!

PCIe 4.0 x1 is sufficient for 10Gbps. You only really need x4 for older PCIe 2.0 cards (which still need x4 even in a PCIe 4.0 slot), but newer ones are PCIe 4.0.

This is one reason I wish they'd release a block diagram. If you look in page 1-8 in the manual, it says it supports four GPUs simultaneously, two at x8 and two at x4, so we at least know that all PCIe slots can be populated at the same time. It also supports bifurcation - the x16 slots can be split into two x8, which is normally used when you want to add more M.2 drives. However, I can't seem to find any information about if any of the PCIe lanes are shared with the onboard M.2 slots. The SlimSAS connector can also be used for PCIe, but again I'm not sure if the PCIe lanes are shared with other PCIe devices. Maybr someone here knows from experience. I have the Micro ATX version of the board, and only two PCIe cards (10Gbps NIC plus a Google Coral AI accelerator). IPMI is a PCIe card for the ATX version and built-in to the motherboard for the Micro ATX version. It's neither x4 nor x8... it's an x1 card that goes in the only x1 slot on the board.

Keep in mind the Asus one has a SlimSAS (SFF-8654) connector that can be used either for PCIe or for 4 x SATA drives, for a total of 8 SATA drives. For a system that you're going to run headless, IPMI is a must-have. My Unraid server is in a closet, and plugging in a screen and keyboard would be a pain.

Most likely. It's annoying though - I don't see other motherboard manufacturers relying on a PCIe card for IPMI. It's still a lot better than not having IPMI, though! I'd guess so too, but they refused my request for a block diagram so I really have no idea how the PCIe lanes are wired nor if any of them are switched/shared across multiple devices.

Can we use our own identity provider or is this restricted to Unraid.net? It would be useful to have this built on top of a standard protocol like OpenID Connect (OIDC) to make it easy to use with our own IdP (like Authentik, Keycloak, Microsoft Entra, etc) - in theory just swapping out the Unraid.net auth URLs with our own.

Unraid 7.2 beta adds support for SSO via Unraid.net account: https://unraid.net/blog/unraid-7-2-0-beta.1. Hopefully the ability to log in using other SSO providers gets added too. This could allow for 2FA support at the identity provider, which would probably be sufficient for Unraid.