There are a few complications here as I had to upgrade to 6.4.0 rc7/8/9 to get my motherboard stable.  As a bit of a backstory:

 

I was running 6.1 for quite some time with a much older socket 775 motherboard but I wanted to upgrade to something supporting ECC and with virtualization support.   After upgrading the motherboard I had stability issues that I troubleshooted for awhile and the solution was a combination of a bios update that the manufacturer originally didn't post (mb_bios_ga-ax370-gaming_f5a.zip) but provided via email, and 6.4.0 RC7 w/4.12 kernel.  I can seem to get it working with up to 12 drives total (trying for 2x2TB and 14x4TB) using some of the on-board along with the SuperMicro AOC-SASLP-MV8 or the LSI 9201-16i but beyond that some random drive will start dropping off.  Typically what happens is when I restart it doesn't seem to detect the drive at all, or it will detect the drive then remove the drive before I can start the array and give something like:

Tower kernel: sd 11:0:14:0: [sdb] Synchronize Cache(10) failed: Result: hostbyte=0x01 driverbyte=0x00
(will be sda through z)

Or if I start using it for awhile it will fill the log with something like:

Tower kernel: REISERFS error (device md1): vs-13070 reiserfs_read_locked_inode: i/o failure occurred trying to find stat data of [4 850 0x0 SD]
(will be md1-md16, currently just trying to get 16 drives going and part with the smaller ones)

And eventually I won't be able to really interact with the array.

Now here's the weird part that I can't quite figure out, it doesn't seem to matter which controller the drives are connected to, which cable, which backplane, or if I remove the backplane and just directly attach the drives with fan-out cables, or swap power supplies, or whatever.  What does seem to fix the issue is swapping motherboard AND downgrading to 6.1 then I can run 22 drives (all I currently have) just fine.  Unfortunately 6.1 doesn't cut it for my motherboard.

 

Anyone know what I can do to get it working in 6.4 rc?  I'm having troubles getting the full log as it's going unresponsive.  I don't think it's related but if I use BTFRS cache pools it also seems to lose the first drive it every so often after rebooting and I have to reformat it to start using it rendering it not usable.  Before I was using the onboard RAID1 with my old motherboard and ReiserFS (because it retrieves accidentally deleted files properly).
 

If I boot the server into Windows, everything gets detected fine and I can read all the volumes with Yareg (albeit read-only and slow) or if I populate it with drives formatted with NTFS it works fine and is stable.  I must admit I am getting a bit sick of formatting drives either because they come up detected as unformatted in unRAID or back to other formats for testing, it certainly takes awhile.  I was having stability issues with some Linux Live CDs I tried so I don't yet have a good test from that standpoint.  Oh and I do have a server Pro license.  It was bugging me because of the "magical" 12 drives but it's the same key file and USB stick I've had since I bought a pair of keys in 2010 (is that old in unRAID years?).  The filename aligns, it does show up properly in the webUI and it does match the original email I received and still have.

AMD Ryzen 3 1200
Gigabyte GA-AX370-Gaming (rev 1.0) F5 BIOS
2x Kingston ValueRAM 8GB DDR4 2400 RAM ECC Micron A DIMM KVR24E17S8/8MA
NORCO RPC-4224 w/120mm fan bracket
Seasonic Prime 650 Platinum (Previously Antec Earthwatts 550)
Matrox G550 PCIe 1x
LSI 9201-16i flashed in IT mode w/20.00.07.00 Firmware & 7.39.02.00 BIOS
Previously SuperMicro AOC-SASLP-MV8 & AOC-SAS2LP-MV8 (seems like it doesn't work with unRAID >  6.1 though, sadly because motherboard doesn't work with < 6.4.0rc7)

 

Pls upgrade mainboard BIOS first. ( At least F6 with Update AGESA 1.0.0.6a, or beta F9a Update AGESA 1.0.0.6b )

 

If problem still presist, I will try plug a non-ECC RAM instead ECC-RAM.

 

I use Ryzen with on-board SATA and LSI SATA, no issue.

Edited October 3, 20178 yr by Benson

The F# aren't contiguous between the different Gigabyte AX370-Gaming series, mine is the plain Gaming version, not Gaming 5 or Gaming K7 so it's version F5a is the AGESA 1.0.0.6b.  It was a key factor to getting mine to run fairly well (not locking up within 2 hrs) and yes I do highly agree it is a must to have it upgraded.

 

Since your post I'm trying with non-ECC, looks like it's still acting funny.. doing more testing though (but getting late for me).

Sorry for overlook different version of mainboard.

 

3 minutes ago, praeses said:

Since your post I'm trying with non-ECC, looks like it's still acting funny.. doing more testing though (but getting late for me).

 

You mean problem presist, i.e. not all drive showup, cache pool show need format etc ....

 

3 hours ago, praeses said:

I can seem to get it working with up to 12 drives total (trying for 2x2TB and 14x4TB) using some of the on-board along with the SuperMicro AOC-SASLP-MV8 or the LSI 9201-16i but beyond that some random drive will start dropping off. 

 

Does no problem if running less then 12 drive, since you change the power supply ( I know bigger then previous one ). But pls note the combine power of 3.3v+5v usually not more then 130w, it was tight for powerup 22 drive (5v part not 12v part).

 

Seasonic Prime 650 Platinum was 100w only.

 

Further check Antec Earthwatts 550, just 90w

 

Edited October 3, 20178 yr by Benson

Ahh..  Thanks.  I was thinking the 12V side for the motor was were most of the current for the drives would be but you may have something there.  I do have a Corsair AX750 (older but working in in another box) that is rated for 25A on the +5V rail.  I'll pull it out tomorrow and see if it makes a difference on the drive count.

Nowaday, that are hard to found a supply which have high combine power of 3.3+5.

Antec truepower 650 is the one which I use and have 170w. But I just try powerup highest 13 hdd only.

 

Edited October 3, 20178 yr by Benson

Tried the AX750, also hacked apart some cables and tested, there isn't much draw on the 5V rail for either the motherboard/pcie cards (30W max) or drives (2-3 watts a piece).  I don't have any free SSDs to test, they may be different.  The bulk is on the 12V rail for both the motherboard (CPU) and drives when they spin up.

 

ECC or non-ECC, doesn't matter.  Both sets pass Memtest without a hitch.  I've got everything in pieces right now and spent more time on this than I can afford so it'll have to be on the back burner for awhile.  Hopefully another version ideally with a new kernel comes out soon to test.  I can't find any fault with the hardware especially since in different combinations it works with older versions of unRAID.  I have tried the iommo switch from:

 

https://forums.lime-technology.com/topic/39003-marvell-disk-controller-chipsets-and-virtualization/

append  iommu=pt  initrd=/bzroot

 

Without success either, as well as disabling IOMMU in the BIOS (or enabling, or default auto).  I'll have to swap back PSUs and get the other computers back together first before revisiting this...

Oops, note with tks

Edited October 4, 20178 yr by Benson