I just dropped the hynix down to 768 today but it was running at 775 for days at a time over the past 2 months and only crashed if I changed OCs without stopping the miner first.
Hynix is using this strap
amdmemtweak --CL 20 --RC 36 --RP 11 --WR 14 --CWL 8 --FAW 12 --RAS 22 --REF 65535 --RFC 239 --RTP 5 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 13 --RCDWR 12
Interesting. Your straps are almost identical, with the biggest difference being -RCDRD (yours 13, mine 12). I believe this setting will give me lower hashrate though, but I will try it as “power saving” mode. You keep your Hynix mems running at 940 though, so that’s another big difference. Maybe that’s what allows you to drop the voltage so low.
I guess I have to run some tests over the next days 
With lower core you might be able to lower vdd too.
Your lower RCDRD of 12 compared to his 13 is indeed one part of your higher hashrate,
but partially also that you have mem at 960 which sets SOC one mem state higher than his 940. Your SOC will be set at 1028 whilst his will be 960.
See: Vega 56 and Vega 64 guide - #679 by oni
And indeed, testing with trial and error is the way to go with Vega’s 
Hello, first thanks a lot for the guide from vsimovic, used his settings and achieved following results
But the question ist why I get still invalid shares on the Samsung Vega 56, it is the newest GPU in my rig, replacing a Vega 56 Hynix with malfunctions.
Do I have to apply other settings for the Samsung card?
Best regards,
Captain
My Hynix mems, set the SoC at 960 whether I have the mem at 960 or 940. Actually, trying to set the SoC manually to a higher state, 1028 as your other post mentions, reduces my mem to 840 and therefore reduces the hashrate a lot (dropped to the 44Mh/s area). So, manually fiddling with SoC, got me worse results than leaving it at default.
Now I am experimenting a bit with memory at 940 (SoC set automatically at 960) and -RC and -RCDRD settings. I want to try to lower my voltage without sacrificing a lot of hashrate. Otherwise I am still happy with my 52.71Mh/s. 
If you like try, on all hynix cards: mem 960 vdd 825 core 1050 with straps, amdmemtweak --CL 20 --RC 36 --RP 11 --WR 14 --CWL 8 --FAW 12 --RAS 20 --REF 65535 --RFC 248 --RTP 5 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 12 --RCDWR 12
If some or all are not working, gpu dies, increase vdd to 830, that should help.
good luck
hey. Tho i was happy with my vega´s, i started to fine tuning all of them…again… This time managed to get wattage down and i was supprised that got one hynix so low W with decent hash…like Oni often says, tuning with vega´s is never done 
These are with eth mode B
Samsung 56 straps
amdmemtweak --RC 37 --RP 11 --WR 13 --CWL 8 --FAW 12 --RAS 22 --REF 65535 --RFC 239 --RTP 6 --RRDL 5 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 12 --RCDWR 12
Samsung 64-56 straps GPU 2
amdmemtweak --CL 20 --RC 37 --RP 11 --WR 14 --CWL 8 --FAW 12 --RAS 20 --REF 65535 --RFC 248 --RTP 5 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 12 --RCDWR 12
Hynix straps
amdmemtweak --CL 20 --RC 36 --RP 11 --WR 14 --CWL 8 --FAW 12 --RAS 20 --REF 65535 --RFC 248 --RTP 5 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 12 --RCDWR 12
Hehe yeah the parameters are so many, that indeed the Vega’s never stop to be tuned.
I have been using your straps with my Hynix Vega for a couple of weeks now without problems, it’s just that I can’t seem to get stability with lower voltages (855mV seems to be the most stable for me - which is quite a lot).
I will try your last settings and see how it goes!
I u feel like having an adventure! …try switch to mode eth mode B. That allows you to lower your core and vdd.
Tune cards with mode A first as good as u can get them, even with a bit high vdd, then switch to mode B and start comming down first with core until u see that your hash dropping–then back up a touch.
Next vdd, down one card at the time like 5 to 10 mV at the time until gpu dies…then back up a bit and your done. Happy mining!..tuning… 
Well I been using eth mode B since day 1 with my Vega! Also been doing this method you describe to fine tune. That’s how I came up to my settings of 1010 Core and 855mV. With 1000 Core I started losing a bit of hashrate, and with 850mV I was getting a random GPU DEAD every now and then (sometimes in 10 hours, sometimes in 3 hours). Ever since I went up to 855mV, I never had a dead gpu again. But it uses so much power and generates too much heat for these hot days. Anyway, will see how it goes with 1000 Core / 800 mV.
Btw, which eth conf number are you using? I find best results with B446
Okay, so you know all this, well maybe the post will help someone else anyways.
I had 488 on all but switched back to autotune, in team red manual it states,
“There generally isn’t any upside from a performance perspective specifying the intensity manually, but for gpus running displays, or gpus becoming too hot, you might want to tune it down to lower the hashrate.”
I have had good results with autotune and no real issues with gpus, well one is running a bit hot, so maybe i need to dial that down at some point if this ambient heat continues. Well see how these values works in the long run, have had now +30hrs without any issues so i think that they might good for now.
Hello again guys. I tried now the timings and clocks mentioned by Smining570 and aminer, as you could see in the picture above, looks good untill now.
And yes you are right tuning with Vegas never stops, btw these are cards from 2017 and still good performing its just insane!
can i know which once are real vega 56 or are these all vega 64 with flashed 56 bios?
all others are 56 only gpu2 is 64 flashed to 56
man your wattage consumption is crazy. well done.
I see you have one Vega56 with Samsung memory. You need to use different timings and clocks for this one. They generally perform better than Hynix, maybe you can reach 55 MH/s. Read a few posts above, @Smining570 has had some impressive results with his Samsung vegas.
Otherwise, your Hynix ones are looking good. You can try reducing your voltage a bit, try 850 first, then 840. Maybe your cards can take a bit lower voltage than mine!
@Smining570 Whenever I tune my cards, I use autotune for a few times and take note of what TeamRedMiner uses. Sometimes it chooses a config that has worse results (don’t know why). Afterwards, I kinda “cycle” these configs manually to see what’s best for me. That’s how I ended up with B446 anyways. It wasn’t all manually chosen.
So, I now have it set to this to avoid the process of auto-tuning that has been done lots and lots of times in the past.
Also, I only have one Vega (sadly). If I had a rig of 4-5 or more, I would probably go with autotune too!
I’ve double checked, and my Hynixes at 960 have their SOC automatically at 1028.
Maybe different biosses set the SOC differently?
These are Gigabytes with bios 113-EXT33736-001.
And the tests I did in that earlier post were with Samsungs on MSI Airboost 2018 bios which were also set to 1028 when mem was at 960.
My Hynix Vega is a Sapphire Pulse with bios 113-376XL-UO2. I haven’t reflashed it, so it should be the stock bios, although I bought it used so I can’t be sure if the previous owner had reflashed it (he said he didn’t, but I can never tell for sure).
So you think that each vendor uses different mem states in their respected bioses?
The only way to know for sure is to flash different biosses on the same gpu and test them all at 960 mem to see at which SOC they are set automatically.
But I don’t feel like flashing them Gigabytes as they have no bios backup like the other Vega’s…
And I lack time lately to fiddle around with my beloved rig 
