It sounds like you have tested the risers one by one connected to the "primary" 16x slot?
The the "only" power issue can be the delivery to the riser, if it has a 6pin connector, use it to take away any converter crap.
If you still have issues, the solution is not to fiddle around with the PSU.
Also, if you only use one card connected to the primary 16x slot the bios could be at default setting since the card will present itself as it was connected to the 16x (this would only test the riser part of the setup).

If you have any mining friends nearby, try to get hold of some riser of another type (preferably one that have worked).

You can also do the homework of figuring ut the PCI bus numbering assigned to you slots on the mobo.
Pressing "s" in Claymore will give stats over the cards and list the GPU0 GPU1 etc in the same order as the PCI Bus number order.

At least in my case (GigabyteZ270PD3 and using SMOS) the PCI Bus number assignment have been stable (but it shifts if you add things as M2 adapters or splitter cards).
After knowing the order of Bus Assignments, you can then route the cables so the cards in the rig is i the same order as the GPU0, GPU1, ...
I was worth the time for me since I know can pull e.g. GPU4 when it kills the whole rig without performing an investigation.

Have you tried other miners than claymore?

I would not think it's ethash per se that is causing you the issues, and if you still get crashes when turning down clocks it's most probably not the heat or power.
That equihash works stable point to no issues with any basic hw (mobo, cpu, memory, risers, GPU memeory/core)...

I would blame Claymore+Windows (drivers etc etc) combination (Claymore 11.6 on SMOS have only occasional misalign issues for me, with 1060 3GB cards).

If you trade immediately it depends on the memory in the 1060.
I have 9 that deliver close to identical 295 sol/s and 24 Mh/s with the same oc settings (140 1400) and max 80w. It's possible to push 310 sol/s but it then takes more power.
There is currently more $ to be made (and more $per W) when using ethash, and there will be several reasons to switch back and fort before the DAG file does not fit.

If your 1060 only can get around 19.5Mh/s, then it's a different calculation.

I run my 1060 at 80w max and mining etherium or monero dos not use all of that at (140 core and 1400 mem), measured trough a 93% PSU (5 cards, less than 400W from walll).
My testrig have happily been running 2x1050 at 50W each and a 80W 1060 using a 450w ancient PSU (290W from wall, since the Mobo is a QC5000M with 15W APU).

Did the AMD folks leave already (or the stampede just stopped)?
https://www.difficultychart.com/monero
Difficulty below 61G now (might just be a correction to the new balance level).

Another Xmr Stak question.

On one of y rigs I get the following over and over again on the monitor (but not in console):

[3680.275839] Failed to expire sync object before unbinding TTM
[3688.601587] [TTM] failed allocating Swap storage
...

Default generated nvidia.txt for the play/test rig (just dual 1050 atm)
...
  // gpu: GeForce GTX 1050 architecture: 61
  //      memory: 1937/1999 MiB
  //      smx: 5
  { "index" : 0,
    "threads" : 32, "blocks" : 15,
    "bfactor" : 2, "bsleep" :  0,
    "affine_to_cpu" : false, "sync_mode" : 3,
  },
...

It delivers to the pool alright, so nothing terminal, 323 H/s each, but could it be a little better without these errors?

Another rig running the same 2.4.2 (same thing was seen on 2.4.1) with default nvidia.txt with 1060 3GB shows no issues.

When you are SSH into the miner and use screen -x miner
you can type c,  h, or r to get connection report, hashrate and results.

No idea if there are command line options to make the hashrate be reported continuously so you can see it in the SMOS console.

This is for 2.4.1 sorry

Yea, as with the server PSUs, compact designs depending on fans going ooooiiiiiiii...

So MoBo designed for for airflow sensitive GPU packing, but with a size that does not fit inside 19" cases...

This line works fine for me, have you tried without the stratus tcp stuff (don't remember if it was due to SMOS or pool if that should be there or not).

--currency monero7 -o europe.cryptonight-hub.miningpoolhub.com:17024 -u bullshituser.$rigName -p x

I got the same error but could remoe that by following the FAQ
https://github.com/fireice-uk/xmr-stak/blob/master/doc/FAQ.md
SSH to the rig,
sudo su
mc -c
edit the files
reboot
-->no error

But in my case the error did not make a difference, it was mining the same before and after (9x GTX 1060 3GB).

/mnt/user/config.txt

Having good experience with a QM5000 board with embedded 15w APU, it would be nice with a barebone solution with maximum width for the GPUs, and very basic extras 8e.g. embedded CPU, DDR3 SODimm etc).

Placing Server PSU(s) on top of the CPU in this case would make it a compact design, but why are the 6-pin connectors sticking out at the bottom of the picture? I would think that would be a case wall in most designs? 19"= 48.26cm MB 48.5cm long ... I.e. it will not fit 19" cases/rigs, so the 6pins at the edge will be sticking out anyway.

Is the heat in these compact designs really a problem? How many "efficient" cards are running higher than 150W nowadays (about 1200W of heat to remove)?
I realize that it would be nice is the failure of one fan does not overheat the whole thing, meaning that a margin (10C?) to terminal failure is needed.
But my knowledge of cramped farms i null. 9x80W in a roomy wooden rig makes the cards run at well below 60C (ok, 19C ambient...).

xmr-stak-v2.4.1-cuda8 (SMOS)
9x 1060 3GB 4923H/s (about 547 each at 80w, 150 core 1500 memory)
I'm probably optimizing it a bit wrong, but it's still more profitable than equihash at the same settings now (cards runs about 7 C colder than on equihash for some reason).

Thought i should share some info on this motherboard after playing a little with the PCI-e, and it fits betetr to continue in this thread that to start a new one.

I use BIOS f8 and the BIOS settings suggested by Vosk, and the M2 adapter i use works.

PCI numbering counting from the CPU side

M2-adapter slot: 10
1st x1: 4
1st x16: 1
2nd x1: 3
2nd x16: 9
3rd x1: 8
3rd x16: 5

But 7 is not enough, so I have played with these "splitter cards" (1 PCI-e to 4 USB connections).
These seem to take 6 PCI-e addresses and shift the numbers of the later slots by 5.

I have added two of these splitters and placed them in the 1st x16 and the 2nd x1 (the ones with the lowest initial numbers). the USB connections is compatible with the v008 risers at least.
This resulted in cards on PCI 3,4,5,6 and 10,11,12,13. And all eight cards are mining happily.
The rest of the slots got shifted from 10,4,1,3,9,8,5 to 20,14,3-6,10-13,19,18,15

I don't have risers and 6pin connectors for more cards right now but 4 more cards to test soon.
It would be nice to know if 2x splitters will make this a 13 GPU motherboard for real (and not only in theory), an alternative when more special motherboards are scarce, but it adds more sketchy components to go bad...

I'm running simpleminer if that matters when it comes to card recognition.
The PCI numbering can be important since it makes it possible to route cables so you have a clue about which card is "GPU3".

Edit:Tested with more risers and a par more GPUs.
Whatever I tried I could not get it to post with 10 GPUs, maximum seems to be 9. with/without the m2 adapter, tried to turn off all I could think off in the BIOS (but too little knowledge to really know what I was doing).

Edit2:
After thinking a bit I thought the splitters was stealing 2 PCI-lanes as it was shifting the PCI address +5 and not +3, and by using two I lost from 13 available down to 9.
So I removed one splitter and used only the normal slots and the m2. Still only max 9 GPUs.
So I removed the m2 and settled fro one 1-4 splitter and the other 5 slots for a total of 9 GPUs (addressed _,9,3-6,8,14,13,10).
If anyone have some info on the ability to disable stuff to be able to add one extra GPU it would be nice to know. Or is it because I'm cheap and use the smallest possible Celeron with only 16 PCI-lanes?