JorgeB

Look for the container support thread:
 

 
 

These usually mean a bad flash drive:
 
May  6 23:50:27 Tower kernel: SQUASHFS error: xz decompression failed, data probably corrupt May  6 23:50:27 Tower kernel: SQUASHFS error: Failed to read block 0x29ae35c: -5 May  6 23:50:27 Tower kernel: SQUASHFS error: xz decompression failed, data probably corrupt May  6 23:50:27 Tower kernel: SQUASHFS error: Failed to read block 0x29ae35c: -5 May  6 23:50:27 Tower kernel: SQUASHFS error: xz decompression failed, data probably corrupt May  6 23:50:27 Tower kernel: SQUASHFS error: Failed to read block 0x29ae35c: -5  

Sorry, missed that, then the issue should be unrelated to binding the GPU, try booting with a new flash drive using a stock install, no key needed, just to see if it boots.

May 6 22:48:27 Tower kernel: BTRFS error (device nvme0n1p1): block=1336104894464 write time tree block corruption detected May 6 22:48:27 Tower kernel: BTRFS: error (device nvme0n1p1) in btrfs_commit_transaction:2466: errno=-5 IO failure (Error while writing out transaction) May 6 22:48:27 Tower kernel: BTRFS info (device nvme0n1p1: state E): forced readonly  
Pool filesystem went read-only, this can be caused by a hardware issue like bad RAM or a filesystem problem, I would start by running memtest, if nothing is found backup and recreate the pool, using btrfs again or try zfs.

Looks more like a power/connection problem, see if disks 1 and 2 share something in common, like a power splitter.

Check filesystem on disks 2 and 3, run it without -n and if it asks for -L use it.

See if it's not the 3.3v issue, Google that and should find many hits.

See parity swap.

If the log tree is the only issue this may help:
 
btrfs rescue zero-log /dev/sdm1  
Then re-start the array

Since the extended SMART test failed the disk should be replaced now.

Do the same for the other pool:
 
unassign currently assigned pool devices for that pool
start array
stop array
reassign all 4 pool devices
start array

Looks like there's something else reading from disk2

The NVMe device is being passed through to the Spiele VM, correct that and it should resolve the issue.

You can use parity swap

Does devices are as if they were just precleared and never formatted, if you still have the original data, format the disks and post new diags after that.

=== START OF SMART DATA SECTION === SMART overall-health self-assessment test result: FAILED! - media has been placed in read only mode  
NVMe device failed and needs to be replaced, you try mounting the pool read-only to see if it work to copy the data.

According to blkid that disk is btrfs, not xfs, click on it and change the filesystem to btrfs or auto, then re-start the array

Various crashing and segfaults, would recommend running memtest before anything else.

Just so that I can mark it solved.

With XFS, the larger the disk, the larger the overhead.
 
It will be same, unless you format with a different filesystem, btrfs and zfs don't have basically any overhead, just a few KBs.
 

Unassign disk1 for now, and also unassign the other cache device, start array, stop array to reset the pool, re-assign both cache devices, start array, the pool should now import, if OK, stop array and re-assign disk1 to rebuild, if not post new diags.

Looks more like this issue:
 
 

Contac support and they can send you a new link to the key.

Both btrfs and zfs are detecting data corruption, and Ryzen with overclocked RAM like you have is known to corrupt data, after fixing that, it would still be a good idea to run memtest.

Btrfs is detecting data corruption, start by running memtest
 
Also change docker network to ipvlan.