Assuming you have your modded roms flashed to all cards correctly.
1. Let's reinstall your drivers:
 - Run amdcleanuputility.exe and let it cleanup your drivers. Once completed. Reboot.
 - Download official latest drivers from Radeon web page. As of writing this post that would be: win10-64bit-radeon-software-adrenalin-edition-18.3.3-march18.exe
 - Install the drivers. If you are only using the GPUs for mining, use custom install and only install driver. There is no need for other components (audio, radeon settings, ...)
2. Update registry (Enable compute mode):
Open RegEdit.exe and navigate and expand:
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Class\{4d36e968-e325-11ce-bfc1-08002be10318}\
Identify all the 0001, 0002, 0003... in my 13 cards rig I have 0000 ... 0014, where 0001 to 0014 are the GPUs and 0000 is some generic.
Now in each of these 0001 ... 0013 create a DWORD 32 bit value named: KMD_EnableInternalLargePage
Set it to value 2.

I do this by inspecting the registry and the updating my template file (Video-ComputeMode.reg) then merging the file:
TO make nnnn a kandidate for GPU it must have some keys included such as: KMD_EnableBrightnessInterface2, KMD_EnableContextBasedPowerManagement
If you do not see these keys, then leave this key alone.
KMD_EnableInternalLargePage will not be in any of them - you have to add it.
-------- START FILE HERE -----------
Windows Registry Editor Version 5.00

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Class\{4d36e968-e325-11ce-bfc1-08002be10318}\0002]
"KMD_EnableInternalLargePage"=dword:00000002

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Class\{4d36e968-e325-11ce-bfc1-08002be10318}\0003]
"KMD_EnableInternalLargePage"=dword:00000002

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Class\{4d36e968-e325-11ce-bfc1-08002be10318}\0004]
"KMD_EnableInternalLargePage"=dword:00000002

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Class\{4d36e968-e325-11ce-bfc1-08002be10318}\0005]
"KMD_EnableInternalLargePage"=dword:00000002
-------- END FILE HERE -----------

This is just an example of the file, you need to update this to match your actual keys present in the registry as described above.

3. Download and run atikmdag-patcher.exe
4. Reboot and start your miner. It will HASH the old values you had in the past (it depends on what you modded to your roms).
For best performance you can also tune the freq/voltage using the Phoenix miner options: For example I use:
-cclock 1200 -mclock 2200 -cvddc 850 -mvddc 850 -- which gives me 31.9 MH/s.
On some MICRON RX580 even -cclock 1220 and -mclock 2220 hashing at 32.100 Mh/s stable in ETH only mining.
Personally I do not use any 3rd party tuning tools such as MSI Afterburner, Wattman ... The miner can set the custom freq./voltages just right.

BTW: The difficulty level is unrelated to the actual hash rate of the GPU.

Should you have further questions or need assistance send a PM message with your Skype ID and I will contact you ...

Good hashing.

Just gave a chance to PM2.8c today on Windows 10. Sharing my first expression, observations, problems and results:

Copied the Claymore config file over. It fails out of the box. Requires a migration of configuration.
After a bit of tweaking a successful configuration file migration was in place. Got the PM2.8c running in less than 3 minutes.

The following discrepancies were observed:

PM2.8c vs CL11.5
1. config file with spaces listing multiple single option values, for example: "-cvddc   835, 835, 850, 900, 835, 835" reports an issue with invalid option 835. CL allows spaces. Some use spaces with a line above as comment with list of GPUs; GPU1 GPU2 GPU3 ... - spaces are for alignment.

2. Order of GPUs is different, this leads to incompatibility with CL existing configuration. CL lists AMD cards then adds NVidia cards. On PH2.8c all are mixed as discovered at system level The migration requires redo of the config file and custom settings for all options with individual values per GPU. For example: -cclock 0,0,0,0,1170,0,0,1160,1160,1160,1180,0,1120 - the 0 for NVIDIA cards. Migrated from -cclock 1170,1160,1160,1160,1180,1120.

3. -ethi default is different 8 vs 12.


Results/comparison as reported by miners (13 GPUs rig: 5x AMD RX580, 7x NVidia 1070):
CL11.5: Average speed (as observed in logs): 400.309 - ranging: 400.151 MH/s - 400.706 MH/s @ wall wattage: 1620W
PH2.8c: Average speed (5 min): 402.817 MH/s - ranging:  400.926 MH/s - 404.945 MH/s @ wall wattage: 1630W

Observations:

1. Can confirm increased hash rate of 0.62%. (Same settings as CL).
2. The reporting API works as advertised out of the box. Using Claymore's Monitor on Android it just works, reporting the version as: PM2.8c-ETH compared to 11.5-ETH (Claymore).
3. It appears stale shares value is at a similar level to CL. The PH reported % of stale shares is not in sync with pool report (ethermine.org): 1.26% vs 3%
4. Log file entries are detailed and easy to understand.
6. The on-screen run log/report is on a higher level compared to CL. Way more informative and rich with data. Not sure about the (!) next to Share actual difficulty when diff is in GH.
7. As a software engineer I do start counting my GPUs with 0, 1 ... but on the other hand some other SWs such as HWinfo64 are also listing GPUs from 1, 2, 3... Not sure yet how annoying if at all this is.

Will keep this rig running for a while to gather more data on stability and performance. Will report back.

Would be nice for miners to add a migration section (from other miner apps) to PH.
Definitely would be interested to buy a licensed version with devfee 0.
All in all a great alternative miner with better performance and a lower devfee. A solid ****. Excellent job.

According to the Claymore documentation and actual tuning results, dcri value has significant impact on performance, check the image from the original post and see how the hash rate changes by setting different dcri value in a range from 1 to 75.

Relevant Claymore Readme.txt snippet:
-dcri   Decred/Siacoin/Lbry/Pascal intensity, or Ethereum fine-tuning value in ETH-only ASM mode. Default value is 30, you can adjust this value to get the best Decred/Siacoin/Lbry mining speed without reducing Ethereum mining speed.
   You can also specify values for every card, for example "-dcri 30,100,50".
   You can change the intensity in runtime with "+" and "-" keys and check current statistics with "s" key.
   For example, by default (-dcri 30) 390 card shows 29MH/s for Ethereum and 440MH/s for Decred. Setting -dcri 70 causes 24MH/s for Ethereum and 850MH/s for Decred.

My newbie label explains it all, and moving the post to another topic can only be done by administrators...
Sorry for spaming the bitcoin...

Thought it might be interesting to post the tuning results of changing the -dcri option on Claymore 11.0 (mode 1 / ETH Only mining).
CardsL 9 x AMD RX 580 8Mb



In this case the dcri tuning tool suggested to use:
-dcri 7,7,7,6,9,8,8,7,7
as best dcri settings for the mining.

I have configured 40+ GPUs achieving 30.0-31.9 on cards with 140-150W on wall per card on ETH only mining.
I use my own approach in dealing with the wiring/bios/modding/running Claymore and setting the undervolting/underclocking values for each card individually. Each card is different so it needs to be tuned individually.

As as see many newbies challenged with the setup I am offering my assistance to anyone willing to match my results.

If you are up for a test drive, I would be willing to help you tune your system directly to a similar results to mine (see below). Done this few times using Skype/TeamViewer - works great.
Send a request via PM.

Next is my current status report from Claymore 11.0 (in total running 1.25GH/s stable):

Rig number #1: 13 cards all ASUS ROG STRIX TOP or OC Edition 8GB with Hynix and some with Micron memory.
ETH - Total Speed: 406.426 Mh/s, Total Shares: 2693(198+209+191+216+211+196+228+203+229+175+202+232+203), Rejected: 0(0+0+0+0+0+0+0+0+0+0+0+0+0), Time: 07:48
ETH: GPU0 31.390 Mh/s, GPU1 31.593 Mh/s, GPU2 30.882 Mh/s, GPU3 31.742 Mh/s, GPU4 31.531 Mh/s, GPU5 31.208 Mh/s, GPU6 31.385 Mh/s, GPU7 30.805 Mh/s, GPU8 31.374 Mh/s, GPU9 31.385 Mh/s, GPU10 31.385 Mh/s, GPU11 31.048 Mh/s, GPU12 30.697 Mh/s
Incorrect ETH shares: none
1 minute average ETH total speed: 405.764 Mh/s
GPU0 t=49C fan=0%, GPU1 t=52C fan=33%, GPU2 t=51C fan=32%, GPU3 t=52C fan=0%, GPU4 t=51C fan=0%, GPU5 t=51C fan=0%, GPU6 t=51C fan=31%, GPU7 t=51C fan=31%, GPU8 t=51C fan=0%, GPU9 t=50C fan=31%, GPU10 t=51C fan=0%, GPU11 t=53C fan=32%, GPU12 t=54C fan=27%

Rig number #2: 13 cards: 9x - ASUS ROG STRIX TOP or OC Edition 8GB with Hynix and some with Micron memory, 4x Gigabyte Nvidia 1070 8Gb (GPU9-12)
ETH - Total Speed: 399.940 Mh/s, Total Shares: 7300(564+614+570+560+563+528+555+548+516+598+544+555+585), Rejected: 1(0+1+0+0+0+0+0+0+0+0+0+0+0), Time: 21:37
ETH: GPU0 30.641 Mh/s, GPU1 30.595 Mh/s, GPU2 30.793 Mh/s, GPU3 30.613 Mh/s, GPU4 30.753 Mh/s, GPU5 30.554 Mh/s, GPU6 30.600 Mh/s, GPU7 30.649 Mh/s, GPU8 29.419 Mh/s, GPU9 31.323 Mh/s, GPU10 31.323 Mh/s, GPU11 31.354 Mh/s, GPU12 31.323 Mh/s
Incorrect ETH shares: none
1 minute average ETH total speed: 399.767 Mh/s
GPU0 t=53C fan=33%, GPU1 t=51C fan=31%, GPU2 t=50C fan=29%, GPU3 t=51C fan=31%, GPU4 t=50C fan=31%, GPU5 t=52C fan=38%, GPU6 t=51C fan=32%, GPU7 t=50C fan=30%, GPU8 t=50C fan=0%; GPU9 t=63C fan=49%, GPU10 t=66C fan=54%, GPU11 t=61C fan=45%, GPU12 t=63C fan=48%

Rig number #3: 13 cards all ASUS ROG STRIX TOP or OC Edition 8GB with Hynix and some with Micron memory.

ETH - Total Speed: 411.677 Mh/s, Total Shares: 11652(895+845+871+858+942+892+896+924+904+942+891+887+905), Rejected: 1(0+0+0+0+0+0+0+0+0+0+0+0+1), Time: 33:36
ETH: GPU0 31.817 Mh/s, GPU1 30.646 Mh/s, GPU2 31.593 Mh/s, GPU3 31.901 Mh/s, GPU4 31.941 Mh/s, GPU5 31.939 Mh/s, GPU6 30.881 Mh/s, GPU7 31.951 Mh/s, GPU8 31.960 Mh/s, GPU9 31.951 Mh/s, GPU10 31.951 Mh/s, GPU11 31.388 Mh/s, GPU12 31.757 Mh/s
Incorrect ETH shares: none
1 minute average ETH total speed: 411.193 Mh/s
GPU0 t=52C fan=33%, GPU1 t=51C fan=34%, GPU2 t=49C fan=0%, GPU3 t=53C fan=0%, GPU4 t=51C fan=0%, GPU5 t=50C fan=29%, GPU6 t=50C fan=0%, GPU7 t=50C fan=0%, GPU8 t=52C fan=0%, GPU9 t=49C fan=0%, GPU10 t=50C fan=0%, GPU11 t=54C fan=30%, GPU12 t=52C fan=30%

BTW: One of my ongoing projects is creating autotune tool to individually tune the -dcri option in Claymore. Will share results/tool when complete.

The clack is when a PSU port line OCP is overloaded and shuts down to protect the circuitry. With the balanced wiring you should be able to run you rig with two PSUs only. How are your risers connected to the PSU? Hint: Connect the risers to the 8-pin ports (as many as you can). For the rest - when you run out of 8 port cables-empty ports on PSU) - connect to the 6-pin ports. Only connect max 2 risers per sata/molex cable.
Identify (User Manual) what are the individual ports max amperage values and make sure you are not chaining too much to the same port or port group.

Background: On my PSU (Corsair AX1500i) there are ten (10) 8-pin port connectors each protected by 40 A OCP and six 6-pin for SATA/peripheral protected all together by 40A OCP. So all 6 ports (6-pin) combined can  only sustain at max level 40 A. This can sustain up to 6 risers. Connecting 7th riser on this same set makes in my case exactly your behavior, which in my case happens in 10 minutes of running the miner, consistently.
When you connect your GPU to the risers and an 8 port power it will draw up to 75 W (about 5-6A). Connecting 10 risers, all to SATA ports would yield 60 A, combined.
In your case I guess you are combining 10 risers to the 6-pin PSU ports via SATA-molex cables and so pulling too much power and the PSU is clacking it self down. This only happens at maximum load when you start mining at highest rate.

The occurrence of power shutdown when an PSU OCP is exceeded is harmful and can burn your components.

If you need more help PM me.
If this solved your problem send your gratitude to ETH: 0x4d45B1B0467e3D02e686936DBE0659e74Bb2Cf52