_cjd_

Not everyone jumped on 7.3 (some can't due to licensing). So 7.2.x is a maintained version line at least for now - and it got the patches before the 7.3.x version got them. 7.3 is a significant feature addition; 7.3.x is patches - no feature change. So "newer" by release date is not the only thing you need to understand. (Unraid appears to follow semantic versioning if you want to go read more on it)

Updated both backup and main servers from 7.3.0 without issue.

It should just work as far as detection. Network setup changes will depend on a few things, but I'd recommend screenshots or otherwise documenting everything on network, as well as docker assignments and VMs.

The pic with the card in the slot you can clearly see it's silkscreen and solder pads. I don't have the board so can't say 100%, but reasonable confidence.

I can't be sure, but if you compare the left side of the slot to the right there's a clear difference (left side is closed for sure), it sure looks like it's open ended. You can see the same white on the left side, and I'm pretty sure that's silkscreen on the board itself.

You don't downgrade, you wait for 7.3.1... I'm sure it's not far behind. 7.3 includes some pretty big changes not everyone wanted to jump on. This is for those that didn't.

Successfully updated both my servers from 7.2.6 - backup server first just to step through the process. just the 7.3 upgrade, made sure everything was working Updated to internal boot (dedicated Optane16gb), auto bios update worked and no issues. only surprise was that somewhere in all my hardware jumping I'd left the fTPM disabled; Have not switched to TPM key yet, may stick with usb, but no issues enabling that and booting still. Main server I was a little more nervous about given the vague comments about VLAN issues and docker issues; The above steps were in part to give me confidence I could roll back in a hurry. Regardless, stepped through one at a time and everything went smoothly. All dockers still working properly, including those on VLANs. No issues with powertop, the same blockers for high c-states still there but still hit c3 as before. Both servers offer the option of switching to TPM for the licensing key. Neither having issues with the dedicated boot. Oh, and no issues accessing the GUI with Firefox 150.0.2/linux - I am using self-signed certs and local dns though, so maybe not quite what others are having issues with. I did verify IP and <servername>.local both work. I did do a bunch of work ahead of the 7.2.5 update due to a history of odd issues - I think the key changes there were disabling "Host access to custom networks" and finding an alternative solution for the ONE docker which I had requiring it, and possibly dropping the custom MTU on my internal 10g lan port to 1500. The latter speed caps at ~8gb/s vs 9.8gb/s with jumbo packets, but ssd speed is the actual bottleneck anyway so I haven't looked back. I do use macvlan as well for dockers, and had to do a bunch of of manual MAC transcribing. Need to add custom MACs to the ones where it matters less, 'cause it royally screws with the network topology mapping and monitoring.

Both ports work with an x8 card in an x4 slot

It sure does.

One port definitely is good on an x4. "eBay card" I mean a server pull, $30-50usd for x710-da2 used. As long as the slot itself is "open" at the back to fit the larger card (or an x8 or 16 slot only filled to x4), it'll run just fine on x4. Obviously you could also do one of the Mellanox cards and single port are available in x4 size (I have those too, single and double port, but the power efficiency is better on the x710 if you chase that). Even at roughly the same cost I'd go an SFP+ card and DAC over the SFP rj45 adapters.

The adapters are as much or more than an ebay card, each. Unless you saturate both ports x4 is fine. I run mine in an x4. Still the latency and efficiency advantages and only iperf can even max it out because drive speed is the wall.

Why not grab an x710-da2 and run DAC instead? Unless it's a cable run problem that's the smarter way to go (and cheaper). Lower power, run cooler... I have two because I've failed every fiber run attempt to two locations... but unifi sfp+ 1/2.5/5/10gbe adapters for my unifi switch. You just need modules compatible with your microtik.

Yeah, drivers is always the fun bit. This is no small task. The Intel stuff is often easier, glad the onboard nic worked! The process was good still, even able to use a blacklisted usb, install to that usb with immediate switch to internal and tpm and never a license alert. The drive is flaky but I tried it thinking the usb adapter was the problem. The surprise was dedicated boot failing. I'd just received the optame drives (one was actually DOA) so it was a bad assumption initially the drives were the problem. In the end the question is, is it worth the effort. For that I'd be curious what this looks like as part of recovery - or if it has to be the existing approach there. That's even more work! I did find it smooth and easy, nicer than finding the latest usb creator image on whatever system I happen to be on.

Took some time today to test this out - I snagged some 16GB Octane drives preparing for internal boot and wanted to explore a few things, verify some hardware and figured why not test this out too. Bug found: I am unable to create a dedicated boot drive directly. The failure seems clear: it's actually trying to create a boot partition with 0 bytes. Steps taken after usb (from .img) booted up: Download image (7.3b2) Set Boot size (Dedicated, 0) Install Not sure of an easy way to grab the log, so the useful bits here with a bunch redacted to save typing: Boot pool target size: dedicated ==> [1/9] Preparing... ==> [2/9] Wiping existing partition table ... calling ioctl to re-read partition table: Success ==> [3/9] Creating bootable target partitions Using DISK_ID: ... Using /boot-transfer writing GPT on device (nvme0n1), with boot size 0MiB ... Creating new GPT entries in memory Could not create partition 3 from 1048576 to 0 ...I was able to set an 8GB partition and install on the same setup, and I was able to install to flash and immediately convert to internal boot on the first boot-up (dedicated drive). Second potential issue: I was getting full screen flash on redraws as it updated progress % - anyone that has a flashing lights sensitivity might be in trouble. 3440x1440 monitor in case that has anything to do with it (no stretch so it detected properly) Another observation: I initially suspected my install media (an old microsd card in a reader) might be the problem so switched drives; on the old/slow drive the initial GRUB screen provides no feedback whatsoever (nor any options) for quite some time - I thought the system had locked! Switched to a flaky usb3 drive and it was a completely different experience. May just be how it is. Finally: I have an Intel x710-da2 in the machine; the BIOS sees it just fine. The installer came up "no network" but it worked perfectly once actually booted into Unraid (used the onboard 1g lan for install).

This misunderstands flash in the first place, but that's not specifically at issue here. The USB stays powered on; it may stay hot, especially usb3 devices. And combined with the writes that happen when changes to configuration are made (possibly syslog if you have that enabled to flash...) and also occasionally issues folks have encountered with ASPM and USB suspend causing corruption, there are in fact several reasons an alternative is quite attractive. So MTBF is never even approaching infinite. My first (Samsung Bar 32GB, in a usb2 slot) died after 3 years (no syslog to flash). I've had corruption issues as well. All recoverable, but incredibly frustrating to chase down. Will eliminating boot from USB fix these issues? That's not a given either, of course (I've had one SSD fail as well) but even the "good" consumer USB drives are increasingly problematic (as others have already noted). And they're not obvious as USB corruption, so I've now replaced all of the hardware in my server (I don't believe I'm alone having gone this far).