Vega 56 and Vega 64 guide

[Smining570]

quote from general usage for TRM "Ethash options:
–eth_config=CONFIG Manual ethash configuration for the miner. CONFIG must be in the form [M][L].
The [M] value selects the mode which can be ‘A’,‘B’, or ‘C’.
The ‘B’ mode uses additional memory and will only work on 8+GB cards.
The ‘C’ mode uses additional memory and will only work on 16+GB cards, such as the VII, with
a correctly configured system. See the ETHASH_TUNING_GUIDE.txt for more details.
The [L] value selects the intensity and it’s range will depend on the GPU
architecture.
Both values are optional, but if [L] is specified, [M] must also be specified.
Example configs: --eth_config=A
–eth_config=B750

                          CONFIG can also be a comma separated list of config values where each is
                          applied to each GPU. For example: --eth_config=A,B750,,A288
                          Any gpu that does not have a specific config in the list will use the first
                          config in the list."

So it should be quite straight forward to set each gpu to different mode and intensity. I havent tried yet, but these extra configs would be handy to set for few cards that are running hot to lover intensity or different mode alltogether.

[CaptainKrypto]

Sry I´ve forgotten to say that my cards are all original Vega 56 with Vega 56 BIOS.

[ss-silver]

Found some new settings yesterday 1027 memory is great! Dropped like 8 watts off my Rog Strix and a 64 on this first rig.

I was surprised how many of the Samsung cards were able to run at 1027 and how low you can set vdd at 1028 SOC level.

Dropped 28 watts off this rig and actually increased hash by .5

• 1 & 4
  amdmemtweak --CL 19 --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
• 2
  amdmemtweak --CL 20 --RC 37 --RP 12 --WR 14 --CWL 8 --FAW 12 --RAS 25 --REF 65535 --RFC 248 --RTP 5 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 13 --RCDWR 12
• 3 & 5
  amdmemtweak --CL 20 --RC 38 --RP 12 --WR 14 --CWL 8 --FAW 12 --RAS 26 --REF 65535 --RFC 248 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 12 --RCDWR 12

On this other rig cards 4 and 5 are on their own psu, it is pulling 270w from the wall right now, so averaging like 135w from the wall per vega on this rig.

• 1 & 5
  amdmemtweak --CL 20 --RC 38 --RP 12 --WR 14 --CWL 8 --FAW 12 --RAS 26 --REF 65535 --RFC 248 --RTP 5 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 13 --RCDWR 12
• 2 & 4
  amdmemtweak --CL 20 --RC 38 --RP 11 --WR 14 --CWL 8 --FAW 12 --RAS 27 --REF 65535 --RFC 248 --RTP 5 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 13 --RCDWR 12
• 3 (this card sucks)
  amdmemtweak --CL 20 --RC 39 --RP 12 --WR 14 --CWL 8 --FAW 12 --RAS 27 --REF 24000 --RFC 239 --RTP 5 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 12 --RCDWR 12

[Austbost]

I have a Vega 56 Gigabyte with samsung memory that I cannot get stable.
Background is I bought this card used along with 3 others, it was flashed to a vega 64 amd and I had it running stable at 38 mh. Fast forward to now and I have flashed it back to 56 gigabyte and cannot get it stable.

Im using the same OC settings as my 3 other Vega 56 gigabyte samsung cards which run perfectly fine at 44mh Core 1030, DPM 5, VDD 815, VDDCI 825, MEM 935 but this recently flashed card gives me a ton of invalid shares then crashes.

Can anyone help?

[ss-silver]

@Austbost If this card has a bios switch try flipping it to the other side first but if that doent help…

Try putting on this strap from the TRM guide, increase VDD to 825, take off the DMP and VVDCI and see if that runs better.
amdmemtweak --RC 38 --RAS 26 --REF 21000 --RRDL 4 --RRDS 3 --RCDRD 12

• Unrelated update on my lower wattage quest
  
  Just hit 10 hours uptime on this, check out GPU 5 55Mh @ 99 watts, also dropped GPU 4 by 9 more watts.
  
  
  
  Ended up using this strap for GPU 5, its a flashed 64 and wont boot with RC 37 over 960 mem.

amdmemtweak --CL 20 --RC 38 --RP 11 --WR 14 --CWL 8 --FAW 12 --RAS 20 --REF 65535 --RFC 248 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 13 --RCDWR 12

GPU 4 is also a flashed 64 but is able to run RC 37 & RCDRD 12 for .8 more Mh and no extra watts.

[Austbost]

Did not find a switch on the Vega 56 gigabyte and tested the strap with no luck, GPU appeared dead

With these settings it will run, but I get invalid shares and will crash every 10 minutes or so

How do I change the DPM settings I cannot find it anywhere even with a google search.

[aminer]

@oni

Is there more info on how TRM chooses that config? Why does it vary?
And I know Smining570 quoted the manual:

@Smining570

“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.”

But why wouldn’t it be better off to set the higher configs amongst the logged ones per gpu for higher hash?
And for the 2 A-mode ones, the lower ones for lower hash and maybe lower temps?

And the most important thing to me; how the hell can I get those 2 bloody Vega’s running on B-mode? Any experiences with this? Should I try to set them manually to a fix lower B-mode config?

I am by no means an expert or guru, but from what I noticed, TRM chooses the config mode intensity based on current “parameters” such as OC settings, temperatures etc.

And also, higher conf number doesn’t necessarily mean more hashrate. Each card has a “sweetspot” I think, and by running the auto-config and taking note of what TRM has chosen each time and then noticing the hashrate and general stability of the cards, I have personally ended up with a “cheat sheet” for all my AMD cards with different eth_config numbers. Using these has never failed me so far, even when I moved cards around between rigs etc. Once you get this “sweetspot” number for each card, then it’s always a 100% consistent result.

Regarding the two cards that can’t run on B-mode, I wouldn’t hassle too much about it, sometimes it is what it is. I have one RX 580 that doesn’t like B-mode, just like your two vegas it crashes when I use it. I even tried switching to B-mode after finding the “sweetspot” number for it (going from A288 to B288). It seemed to work but I got worse hashrate and it would still crash randomly every now and then, so it’s back to A mode. Works like a charm for weeks, so I really don’t bother much.

[aminer]

@ss-silver You made me wanna try and fiddle with the core voltage these days. This is what I achieved with my Sapphire Vega 56 with Hynix memories. I really don’t know how in the past this card could not go with lower than 850 mV.
Now it is stable at 780 mV and only 99W in the software (I assume around 125W at the wall).

[ss-silver]

@aminer haha nice! Ya im trying to figure out why some cards are able to go so much lower. Maybe TRM 0.8.3 had something that helped, I dont remember reading it in the release notes but ya I am now able to run some of these cards 50mv lower than before and looks very stable to me, 31 hrs with thes vvd settings so far and still running at 100%

@Austbost that’s strange
My next thought is to try increasing rcdrd with:
amdmemtweak --RCDRD 15

It will drop hash but if it fixes the invalids you could try adding --RC 37 --REF 65535, which should bring the hashrare back up a bit.
The little blue circle in the image above the individual OC setting buttons is how you can change DMP state for all cards.

[CrazyMoyo]

Thank you vsimovic

Am sharing my results as well. I have one Asus Vega 56 with Hynix and a Vega 64 with Samsung. Both are on stock bios. Using TRM with --eth_config=B

Straps used for Vega56
amdmemtweak --CL 18 --RC 34 --RP 13 --WR 11 --CWL 7 --FAW 12 --RAS 23 --REF 17000 --RFC 164 --RTP 6 --RRDL 4 --RRDS 3 --WTRL 4 --WTRS 4 --RCDRD 23 --RCDWR 11

Straps used for Vega64
amdmemtweak --CL 20 --RC 44 --RP 14 --WR 14 --CWL 8 --FAW 12 --RAS 30 --REF 17000 --RFC 249 --RTP 5 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 14 --RCDWR 12

[oni]

@Austbost

Did not find a switch on the Vega 56 gigabyte

Gigabyte Gaming OC indeed have no switch so no backup bios so be carefull if you flash to pick the correct bios from the first time!

@Austbost

tested the strap with no luck, GPU appeared dead

I have no idea why, but in my experience, Gygabyte Gaming OC Hynix and Reference Card Hynix need different straps, for me at least. I have these set:
Gigabyte Hynix:

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 12 --RCDWR 12

Reference Hynix:

amdmemtweak --CL 20 --RC 36 --RP 13 --WR 14 --CWL 8 --FAW 12 --RAS 23 --REF 65535 --RFC 239 --RTP 5 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 16 --RCDWR 12

(if lots of invalids; increase RCDRD; if non, try to lower it step by step until you see invalids popping up)

PS: Not my straps; all credits go to @ss-silver and @Smining570 ; they’re the gods of straps

@aminer

And also, higher conf number doesn’t necessarily mean more hashrate. Each card has a “sweetspot” I think, and by running the auto-config and taking note of what TRM has chosen each time and then noticing the hashrate and general stability of the cards, I have personally ended up with a “cheat sheet” for all my AMD cards with different eth_config numbers.

Thanks, I’ll take more notes and find the sweetspot!

@aminer

back to A mode. Works like a charm for weeks, so I really don’t bother much.

It wouldn’t bother me if they wouldn’t run a few °C hotter than the other ones, and I’m trying to get their temps down so they can keep running until higher ambient temperature.
I have now 3 templates with different straps and OC’s; and these templates keep the HBM’s to max 70°C. Lower straps/OC’s mean lower hashrate, but if the longevity of the gpu’s is better than I’m glad to sacrifice some MH/s.
On my 13GPU Vega rig (5x Hynix, 6x Samsung 56, 2x Samsung 64>56) this gives about the following results:

T<23°C A-mode Hash 703MH/s HiveOS 1762W Wall 2430W
T<23°C B-mode Hash 702MH/s HiveOS 1675W Wall 2330W
T<27°C B-mode Hash 674MH/s HiveOS 1535W Wall 2140W
T<31°C B-mode Hash 658MH/s HiveOS 1480W Wall 2080W

The difference in hashrate between the last 2 is negligible, so I’m usually skipping the <27 one and use the <31 more often.
And I lack time to play with it lately, so I’m sure I can do better finetuning than it is right now, but it’s satisfying for the current temperatures. If summer hits >31°C I’ll start sweating again, but not for ambient temps but for downtuning he rig again…

The straps and OC’s are from @ss-silver
<27 straps and OC’s Core 1010 Mem 960
<31 straps and OC’s Core 960 Mem 900

[oni]

@Austbost

How do I change the DPM settings I cannot find it anywhere even with a google search.

I think that’s a bug in HiveOS; if you have ever had other gpu like rx580 in the same rig and had set DPM state for that one, then replaced the gpu by a Vega, the DPM state is still there.
You can do 2 things:
Restore that other gpu and remove it’s DPM setting, and restore the Vega afterward.
Or more easy, note down all your other OC’s (you have a print screen here already), then go to the Overclocking TAB and click the Reset All button. This will remove all OC’s so also the DPM state. After that, just enter the OC’s again which you have noted or print-screened.

[oni]

@ss-silver

update on my lower wattage quest

Holy *** that’s a nice finding!
I truly love your experiments!

@Smining570

So it should be quite straight forward to set each gpu to different mode and intensity.

Yes, that is indeed pretty straight forward; but what makes decide TRM to choose e.g. 448 one time, and next time 472 or 592 as the gpu straps and oc’s didn’t change. @aminer also mentioned temperature; but it would still be interesting to know what influences it as the temperature is the only thing I can’t control (entirely).

[ss-silver]

haha I am just glad someone enjoys all the pictures I post

Working on my “worst Samsung vega” today, and found something interesting…
GPU 3 here has been throwing invalids with every strap i’ve tried and all the way down to 960 mem, went back to the strap from the TRM guide and hit 24 hours with no invalids at 960. Slowly started changing each setting from the TRM strap to the levels I run on my best cards, and found what was causing the invalids was the RRDL?!

Stock Samsungs have a RRDL value of 5, TRM guide suggest turning it down to 4 but most straps have it set to 6 since setting this value to 6 increases hash by about 0.1- 0.2Mh

after 12 hours at 100%, changed RRDL from 4 to 6 and got 2 invalids in 90 minutes, using this
amdmemtweak --RC 37 --RAS 20 --REF 65535 --RRDL 6 --RRDS 3 --RCDRD 12

stress testing that value by only changing RRDL in my normal strap and pushing mem to 1027, 3 hrs so far at 100%
amdmemtweak --CL 20 --RC 37 --RP 11 --WR 14 --CWL 8 --FAW 12 --RAS 20 --REF 65535 --RFC 248 --RTP 5 --RRDL 4 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 12 --RCDWR 12

[angryMiner]

Hello guys!

This is my AMD rig:

GPU 0 - Reference Vega 64
GPU 1 - Asus Vega 64
GPU 2&3 - Gainward Vega 64
GPU 4 - Gigabyte Vega 56

I’m ending the set up of the straps, so I dont post them yet, but there’s nothing special, just step by step and slow work. RRDL at 5 for all cards up to now.
I’m really enjoying this kind of job, Nvidia 3000s are more much boring! lol

But the only I don’t understand, its why there’s that big difference of temperature between both gainward cards, similar power consumption, and new thermal pads for both… 11 celcius is a world!

Any idea?

[angryMiner]

And then there’s the ICU…

This card is not working fine, so it was mining @ 44MH/s with no straps for the moment. With ss-silver straps now it’s mining @ 49MH/s and still stable.

amdmemtweak --RC 38 --RAS 26 --REF 21000 --RRDL 4 --RRDS 3 --RCDRD 12

It’s a reference GB Vega 56, and fan won’t work over 40%, so it gets so hot. I think I will try to connect the fan to the motherboard or molex, because with better temps i would probably be able to put decent straps.
This is happening since i clean up this oxide:

Hope dont go worse.

So thanks for the straps =D

[Smining570]

i have on my 3 ref 56 same oc on all, with proper cooling i can run them mem1020 vdd815 core 1100
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
These gives around 55.4mh

Did one new ref card´s paste, i didnt had in hand any pads, only 1,5mm pad for the small chip that is a bit low on the board.

There was still factory seals, a virgin, thermal paste was hardened really bad, shame i didnt had 1mm pads, will open it again later. So after thermal paste and 2 small thermal pads, it was running with core around 62c and mem 73-4, now core 52c and mem stable 71c.
sneak peak to mining chambers if u want, take a look.

Show us your rig or rig´s!

[oni]

@ss-silver

RRDL

Thanks again for this great finding!
For the purpose of testing, this morning I also lowered it to 4 on my 2 Vega’s that kept crashing in B-mode and you know what? Now already 4h30 up and counting!!
After 24h test I’ll start lowering their clocks and mems too so their temps decrease to the same level as the others

Must test that 1027 too one of these days

[oni]

@Smining570

Show us your rig or rig´s!

Nice!
I’d rather not post on YouTube myself, but maybe we could start a thread somewhere in the hive forums if it doesn’t exist yet?
Also curious about all your setups

[oni]

@angryMiner

But the only I don’t understand, its why there’s that big difference of temperature between both gainward cards, similar power consumption, and new thermal pads for both… 11 celcius is a world!

Any idea?

When repasting the GPU/HBM I have noticed that when remounting the cooler, you have to really really tighten the screws or there can indeed be a big difference. I haven’t had as high as 11°C though, but had up to 6 to 7°C in the past which disappeared after repasting again and tightening.
Don’t know if your issues are related, but might be worth investigating.