I also got 6-6.2 on my 1080Ti's, experiment with -i

I think the problem is the lack of knowledge of the principals of pool mining.
It would be quiet long to describe everything here.
First you have to know what a hash function does, then it will make sense.

difficulty: simply said it's equivalent of the number of the leading zero values in the hash result. More zeros mean higher difficulty, because the set of valid values is decreasing with more zeros (like a shrinking target when shooting)

share: a solution proposal for the current block (nounce is the key here, DYOR)

worker difficulty: the difficulty limit of your pool worker, if you find a share with a difficulty higher than that, it's sent to the pool.

network difficulty: the difficulty calculated from the hashrate of the entire network.

share difficulty: the actual difficulty of the found share. Only a share with a difficulty higher than the network diff is accepted as a valid block candidate.


So for you question (whats your share diff) the answer is that share diff is calculated straight from the leading zeros of your found hash.
The miner program only shows shares that has difficulty higher than you current worker diff.
If you are mining in a pool that's using dynamic worker diff you can see a lot of shares in the first few minutes while the pool is adjusting your worker diff according to your hashrate.

yeah, nothing special, too much oc and/or too low voltage and/or bad timings

You simply cannot compare PL % settings of different cards.
Percentages don't tell shit, because it is relative to the TDP set in the BIOS of the particular card.

Card A has TDP 250W (average 1080Ti with AC) and card B has TDP of 280W (1080Ti with LC).
70% means 175W for card A and 196W for card B.
It's obvious that card B will perform faster (if the clock/voltage curve is similar, but that also differs adding more complexity to the problem)

The same logic also applies for Core and Mem clock offsets. +100  means nothing, because 0% GPU clock can be anywhere between 1500 and 2000MHz.
So please stop comparing them among different cards, makes no sense.
You have to compare the exact clocks.

No. 5V is needed for risers with molex 4pin connector.
You have to use 6pin risers that only need 12V.

just my 2 cents

For mining server psus are the best, they are robust and built with better components than the most expensive ATX PSUs.

12 cards are working fine for me.

It doesn't matter if they use HCS connectors if the wire is still 18AWG that can handle only 7Amps max.
Just my 19.9 cents.

My other remark is that the connected pins differ card by card. AMD likes to connect all sense pins also to GND on RX series so they have 3pins 12V and 3pins ground in a 6pin connector, making  3x7Amps×12V=~250Watts.

But you should check with a multimeter first where the pins are connected.
I measured some myself.
Nvidia uses the sense pins only for signaling more frequently.
Thats why nvidia cards always need both connectors plugged in to operate.

Very simple solution for that is that you can use as many molex splitters as you want, since there is absolutely ZERO power draw from the molexes on the mobo if you are using powered USB risers.
So simply plug a molex splitter or a SATA power to 2x molex splitter on your PSU and plug the 2x molexes in the mobo.

Don't do this.
You can clearly see that HW error is way too high. 6.2% !!
HW error is acceptable up to 0.5%, above that your effective hashrate decreases significantly and you kill you hw.
I have much experience in this matter I've been running and repairing ASIC miners since 2013.
Also keep in mind that VRM's are not designed to handle 30% more power even if the ASIC could handle 30% higher frequency.

My performance tests showed that almost all L3+ starts having too much HW errors above 462MHz. So 600MH/s is the top.

No there is an issue with the driver itself, but there is a workaround for 6+ GPUs (driver reset with devcon.exe).
My 6 GPU Vega rigs running fine @ 11700h/s @ ~1080W

Most likely caused by temps climbing up.
Increase min fan rpm to 3500 or more if in warm environment.

The problem is that you can never compare power consumption for different Tis in means of PL percentage, because PL 100% is based on TDP which is different for every submodel.
For instance Zotac AMP at 85% needs the same power as an FE at 95%.
Because Zotac has 300W TDP by factory and the FE has 275W and 300*0.85 ~= 275*0.95
(Just an example I just guessed the numbers, but you get the point.)

Of course ASIC quality and voltages also matter. For nvidia asic qual is hidden from outside, but the bios controls the voltages according to the actual chip quality like EVV in AMD cards.

I've also seen an asrock h110 unstable when the x16 slot were in use (through x1 riser).
Strange...

After hours of troubleshooting I found that my asrock h110 pro btc becomes unstable when I use the x16 slot even with USB riser.
I plugged all cards into x1 slots and all is working fine since then. Try to leave the long pci-e slot blank.
There is no such problem with the older pro btc mobo.

probably all. including me too.

I once had 5/6 hashing at around 1800/card, but I couldn't repeat that ever again.
I'm back on normal driver now on ethash, I got fed up with the buggy BC driver.

BTW I could turn on HBCC for all 6 cards, but only 4 of them had the slider.

You probably found the resource allocation limit of the mobo. Nothing much you can do about it without modded MB BIOS.
BTW 13 GPUs is nice achievement.

You have to solder a cap, a transistor and a resistor on it to have a slightly delayed PS_ON. I've seen it some time ago. Serach for it. It works, I tried it.

I heard that the AMD Fire Pro 17.Q3 driver already has the DAG size fix.
You can install it for RX 4xx/5xx without a problem AFAIK.

Yes, the reported HR halved since yesterday.
Maybe they doubled the share diff to reduce server load, but forgot to adjust the calculations.
What most important is the "mined in last 24hr" value. If it's OK, then it's fine.