Flashed all cards back to original bios....
There's something funny going on with wattman.

I have a 1920 threadripper with 12 cores and hashing with 11 (1 core to keep system responsive)
I have Large Page Supoort Enabled and hash at around 90 per core = ~1000

Wattman settings for the vega56's    945 mem / -30 core  - 20 power
(Got it to mine at 1080 mem once with one card at 1.000 volt mem and it hashed nearly 2000 ?!?! then it crashed after 15 secs.)

if i turn on HBCC it does mines at 900+880 = ~ 1780 per card
Stak settings 1522+1512
The card with the monitor attached is set to 900 mem to keep things stable...

All nice and well , rig is hashing around ~7500 with threadripper and 4 x vega 56's @ ~950 watts.

Point is, when i reboot cards hash @ around 550+550 = 1100. And everything in wattman looks the same.
But for some reason HBCC has changed and doesnt show any  changes  in wattman.
Settings look exactly as before rebooting.

If i adjust the slider , (any way) , it starts hashing @ 900 + 880 again.

So wattman does not correctly set HBCC by itself after reboot, and have to set is for all 4 cards again each time.

 i  have 32gb quad channel mem and 70.0000 swap file on fast NVME SSD.

First have to start stak cpu and then stak gpu , otherway around it bugs / hangs...

Any tips and tricks ?

@rednoW

[/quote]
flash your Vega56 with Vega64 liquid bios + apply reg patch. You'll be able to run mem 1100mhz
[/quote]

Flashed a : Sapphire.RXVega64.8176.170719.rom in all 4 cards.

Mem does reach 1000 , but power draw is huge (psu overload)
Did reach 900+900 = ~1800 on -20power -30 core
From 875 watts total with 4 cards to 950 with only three cards.... idle system uses 270 watts.
Per card from 150 => 230 watts
Had to reinstall the driver cause of crash, PITA that it auto enables crossfire on all 4 cards (initialising takes ages...) before it can be switched off..
Did make a reg patch with the xcel tool but the havent reaaly dug into the voltages yet, so maybe not the best patch i have...
Couldn't really continue cause card with monitor seems highly unstable , even at second (vega56) backup bios ... hope card is still ok. will flash it back to standard bios and leave it be as it was originally.

Any reg patch file you can send / point me to / recommend for the other 3 cards with lowest power draw ?
 
(i need to get this rig to work with 1200 psu...)

thanks !

@rednoW

Thanks for the info , didnt really help though.

1. Set you Virtual memory size to 64gb ... >>> done
2. Use blockchain 23Aug drivers...  >>>yep
3. Move HBCC slider to some new position or just switch it off / on depending on your current setting. Hit apply >>> All @ 170040mb
4. Set your HBM2 mem speed to 1100 ?!??!?!?

Its a vega 56 not 64 as i understand 56 has lower clocks (but better timings ? @ 920).... anything over 950 result in a crash
920 seems stable, 950 stability depends on STAK settings

(did the -30 speed / 950 mem / -20 power setting)

5. Run xmr-stak-amd with 2 threads and intensities 2016 and 1800. Maybe decrease intensity for monitor attached card.

2016 / 1600 works ... 1800 crashes...  had it upto 800 once now it's between 600 and 700 no matter what i try it wont go higher

6. (optional) Apply reg patch if you want to lower power consumption. For monitor attached card you will need much higher voltage.
Is the reg patch also working for vega56 (thought is is for 64) but first need to try to get things stable i guess...

P.S. Don't use GPU-Z. For monitoring use latest HWINFO64 with GPU I2C support disabled in settings >>> ok , thanks Both seem to work though

Hi all, newbie in town ! Wonderfull thread btw..

Just build custom EK loop :

threadripper 1920 (with ek tr4 block)
4 x vega 56 (4 x ek nickel block)
samsung pro evo 512 m2 ssd (3200r / 1800w)
gskill 4x8gb 3200 cl 14-14-14-34
Corsair  1200i PSU
480 monsta rad
EK revo d5 xtop dual pump

POIDH : http://i68.tinypic.com/w7oaar.jpg

Max temp is @max 48 degree @ full load... (1000 watts) (normally around 45 water temp / 22 ambient )

Threadripper working with stak cpu at ~950 hashes
Did get the vega's once hashing at ~ 800+800 per card but messed up settings i guess (too many crashes)
now doing only ~200+200 per card ;(

Q.
I have for each card a HBBC slider ranging from about 11000 to 24000.
The HBBC is per card right ?
So 4 cards is minimal 1100 x 4 = 44 GB of ram ?

Q.
One card has monitor attached , it Always seems to be unstable (need lower settings or whatever) , same on my r9 rig i had.
Does disabeling Aero etc. help ?

btw, i see people with loops and ek blocks showing temp diff > 10 degree between hbm and core...
mine is only 1 degree off between the two !
FYI : I did use the Original four legged springy rhingy clamp from the air cooler (just use two plastic washers to keep them higher off the pcb so they dont touch the board. Improved pressure and core contact with waterblock

Q.
GPU z says 4 gb vram in use ?!? (with two thread setting 1012+1012) ?