AMD Mem Tweak XL - Read/modify timings/pp/straps on the fly

[djukica666]

guys what is the danger of keeping tREF to high? I have seen somewhere that the HBM temps might go critical with it to high, but I'm not mining with my vega56@64, only gaming. now during winter with room temps 20-23C I'm not going over 63C on my HBM during heavy load.
these are my latest settings that I managed to get stable under tests/benches. I've been playing with this for the whole weekend. anything more than 1050mhz on HBM freq I get artefacts and crashes, so I left it there and went tightening the timing.

https://postimg.cc/5XR6nhG0

I'm a little disappointed cos I expected more out of samsung HBM when flashing v64bios and going higher on the voltage. I barely get the same results in tests/benches as with default bios and OC/UV/tightening:

https://postimg.cc/qhdBF4Gk

the card is revision C3

thanks.

[1675739303]

1675739303

[1675739303]

1675739303

[1675739303]

1675739303

[1675739303]

1675739303

[percy_tc]

Is there any update on 5700 mem timings?

[pbfarmer]

Hello guys !

finally, I can't do it.. I'm going crazy, i'm giving up !..

Can't reach a right profil without "Incorrect ETH Shares" and same without tweak..

I don't understand how is it possible... I've got 6 Sapphire RX Vegas 64 Nitro+ and I cannot have a profil around 50-51 Mh/s without errors...

I've already try this :

Conf 1 :

GPU :

- P0 : 852 Mhz / 800 mV
- P1 : 1175 Mhz / 900 mV
- P2-7 : Disabled

Memory :

- P3 : 1100 MHz / 900 mV
- P0-2 : Disabled

-> Incorrect ETH Shares

Conf 2 BIS :

GPU :

- P0 : 852 Mhz / 800 mV
- P1 : 1075 Mhz / 850 mV
- P2-7 : Disabled

Memory :

- P3 : 1090 MHz / 900 mV
- P0-2 : Disabled

-> Incorrect ETH Shares

Conf 3 :

GPU :

STOCK

Memory :

STOCK

-> Crash then restart !

Conf 4 :

GPU :

- P0 : 852 Mhz / 800 mV
- P1-7 : Disabled

Memory :

- p3 : 1107 Mhz / 837 mV

-> Settings won't apply in Claymore (I've got 44 Mh/s)

WTF is that ? I don't understand what's wrong ...

My config file with Claymore :

setx GPU_FORCE_64BIT_PTR 0
setx GPU_MAX_HEAP_SIZE 100
setx GPU_USE_SYNC_OBJECTS 1
setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_SINGLE_ALLOC_PERCENT 100
@echo off
cd %~dp0

"EthDcrMiner64.exe" -epool eu1.ethermine.org:4444 -ewal my_wallet -epsw x -eworker my_name -etha -1 -erate 1 -estale 1 -mode 1 -r 0 -tt 59 -tstop 80 -showpower 1 -showdiff 0 -platform 1 -y 1 -strap 4 -dcri 1,1,1,1,1,1

If someone can help me to reach a good profil it very helpful or i'm gonna die haha

In case you're still looking for solutions, here is my actual current setup for ethash, using TRM (autotuned):

amdmemtweak settings for all 8 GPUs: --RAS 30 --RCDRD 14 --RCDWR 6 --RC 44 --RP 12 --REF 15600 --RRDL 4



Bottom right pane shows profiles of 2 GPUs.  First profile can generally be run at 800mv for 47.5MH/s, and is *slightly* more efficient, but not by much.  Second one is for one of the GPUs doing 50.7MH/s - the only thing you should have to change from this is the mclk.p3 voltage setting - i need anywhere between 806-837mv across 7 GPUs doing ~50.7MH/s. Both profiles are using cclk p1 / mclk p3 (although in windows you may need to use mclk.p2 @ 1028MHz for the more efficient profile, to take advantage of the lower socclk - i'm not certain.)

Note this is slightly different from what I mentioned earlier... I was using core p0 (850/800) before TRM - and possibly before amdmemtweak (which likely would have resulted in 44MH/s.)  With TRM + amdmemtweak, it appears you need to have core running ~1000MHz to take (nearly) full advantage of the mem bandwidth @ 1107+adjusted timings (and get to 50+MH/s), so I use core p1 @ 1000MHz + mem p3 @ 1107MHz.

[pbfarmer]

guys what is the danger of keeping tREF to high? I have seen somewhere that the HBM temps might go critical with it to high, but I'm not mining with my vega56@64, only gaming. now during winter with room temps 20-23C I'm not going over 63C on my HBM during heavy load.
these are my latest settings that I managed to get stable under tests/benches. I've been playing with this for the whole weekend. anything more than 1050mhz on HBM freq I get artefacts and crashes, so I left it there and went tightening the timing.

https://postimg.cc/5XR6nhG0

I'm a little disappointed cos I expected more out of samsung HBM when flashing v64bios and going higher on the voltage. I barely get the same results in tests/benches as with default bios and OC/UV/tightening:

https://postimg.cc/qhdBF4Gk

the card is revision C3

thanks.

Given that tREF controls how frequently the memory is refreshed (to overcome charge decay,) the main issue w/ a too high tREF is you will encounter data corruption as bits start flipping from not being refreshed frequently enough (this effect is accelerated by heat.)  And since a refresh is a series of reads/writes, i would think it should not be that different from other workloads as far as the memory is concerned - if anything the core will likely be working harder as it probably doesn't have to do much during a refresh.  It is possible that a refresh is easier on the mem than your standard gaming workload due to r/w size/frequency or something like that, in which case, yes, the mem would draw more power and generate more heat as it does less refreshing and more 'real' work.

As for the 64 bios flash, it's value is really dependent on the work profile of your application/game.  For example in the mining scenario, ethereum responds well to higher bandwidth, meaning higher clocks, which require more voltage and relaxed timings (which happens to be exactly what the bios changes.)  Other coins/algorithms may respond better to lower latency, which means tighter timings possibly requiring lower clocks (and possibly less voltage,) making the bios update less useful.

I believe games in general like the higher bandwidth, so I would think the flash would be helpful, but it's probably a more delicate balance between latency and bandwidth than many mining scenarios.

EDIT: Also, in case it wasn’t already apparent, your higher mem clock (vs stock) will likely require higher core voltage for the same cclk, because the infinity fabric, which shares power with the core, runs at (roughly) the same speed as the memory.  So the increased memory voltage resulting from the bios flash isn’t the end of the story...

[shostt58]

1st test mod bios (straps) rx5700

hex editor, tl866
1500 to 2000 mhz
gpu 0 mod, gpu 1 orig bios

atiflash block flashing, need help about this
drivers block high frq timings (1750-1890 - hashrate about 54mh)

https://youtu.be/nhIiZyTPv94

may be help me, and help i

[biospb]

flashing 5700 with custom bios is not a problem any more  
the only problem is bios editor for Navi... hexeditor is our friend and shostt58 

[Eliovp]

flashing 5700 with custom bios is not a problem any more  
the only problem is bios editor for Navi... hexeditor is our friend and shostt58 

Please define yourself.

AFAIK, flashing a modified vbios (except then flashing an XT vbios to a non XT card) does not work due to the on board PSP basically saying "Nope".
Using a custom softpowerplay table, that is of course possible.

So not sure what you're trying to say here.

[Goulamouk]

Hi every one.

With my Vega56 original bios and Amdmemtweak : 50Mh/s on Etash
Question : Amdmemtweak works with Radeon VII ?   

[argominer]

Hi every one.
With my Vega56 original bios and Amdmemtweak : 50Mh/s on Etash

Which settings and timmings you got stable?

[biospb]

So not sure what you're trying to say here.

that flashing custom bios is possible

[scryptr]

So not sure what you're trying to say here.

that flashing custom bios is possible

NOTHING CONCLUSIVE--

I tried reading the several threads that I found after searching earlier posts in this thread concerning user-modded 5700 series AMD graphics cards.  The user claiming to be attempting BIOS mods and uploading them with ATIflash has poor English or is deliberately using LOLcat-speak.  Nothing posted is legible enough to make sense.

Other long-time gurus explained that that the latest AMD cards have encrypted and signed BIOS. The Navi series won't accept a BIOS flash the way Polaris and earlier cards will.

If you can point me to a legible post or review explaining how to alter a Navi BIOS and flash it, please do.       --scryptr

[Iamtutut]

So not sure what you're trying to say here.

that flashing custom bios is possible

NOTHING CONCLUSIVE--

I tried reading the several threads that I found after searching earlier posts in this thread concerning user-modded 5700 series AMD graphics cards.  The user claiming to be attempting BIOS mods and uploading them with ATIflash has poor English or is deliberately using LOLcat-speak.  Nothing posted is legible enough to make sense.

Other long-time gurus explained that that the latest AMD cards have encrypted and signed BIOS. The Navi series won't accept a BIOS flash the way Polaris and earlier cards will.

If you can point me to a legible post or review explaining how to alter a Navi BIOS and flash it, please do.       --scryptr

Still, AMD is releasing new bioses for the 5600, they raised frequencies and TDP allowance after Nvidia cut the RTX 2060 prices. This means the Navis can be BIOS flashed. Modified bioses is another issue.

[Branko]

Still, AMD is releasing new bioses for the 5600, they raised frequencies and TDP allowance after Nvidia cut the RTX 2060 prices. This means the Navis can be BIOS flashed. Modified bioses is another issue.

Of course they can...even before 5600, you could flash 5700 with 5700XT bios...but all of those are AMD-made, already signed
bios-es. Flashing is not a problem, but signing

[biospb]

NOTHING CONCLUSIVE--

I tried reading the several threads that I found after searching earlier posts in this thread concerning user-modded 5700 series AMD graphics cards.  The user claiming to be attempting BIOS mods and uploading them with ATIflash has poor English or is deliberately using LOLcat-speak.  Nothing posted is legible enough to make sense.

Other long-time gurus explained that that the latest AMD cards have encrypted and signed BIOS. The Navi series won't accept a BIOS flash the way Polaris and earlier cards will.

If you can point me to a legible post or review explaining how to alter a Navi BIOS and flash it, please do.      
--scryptr

Of course they can...even before 5600, you could flash 5700 with 5700XT bios...but all of those are AMD-made, already signed
bios-es. Flashing is not a problem, but signing

search for atiflash293plus

[datspike]

So.. I guess we'll see 5700 bios modding in the future
https://www.overclock.net/forum/28308884-post445.html
https://amia.io/2020/01/03/rx5700-mod-bios-test-eth-mining/
http://dr-shost.com/files/atiflash293plus.zip
http://dr-shost.com/files/rx5700-orig-mod-bios.zip

[Branko]

So.. I guess we'll see 5700 bios modding in the future
https://www.overclock.net/forum/28308884-post445.html
https://amia.io/2020/01/03/rx5700-mod-bios-test-eth-mining/
http://dr-shost.com/files/atiflash293plus.zip
http://dr-shost.com/files/rx5700-orig-mod-bios.zip

Since you're promoting cloud mining which is scam by default, I won't
try other links either, as I suspect those tools probably steal my crypto or
install viruses on my PC

[datspike]

Since you're promoting cloud mining which is scam by default, I won't
try other links either, as I suspect those tools probably steal my crypto or
install viruses on my PC

Bruh, chill man.
I don't know shit about mining, I'm just an overclocking enthusiast.
You're accusing me about something related to scam because of the fact that the source for this guys comment is that amia.io website?

I've tested the bios edits and the timings indeed can be changed on the bios level. Look here - gpubiosbase reports 1750 straps at 1875/2000 memory clock for the edited bios.
https://i.imgur.com/lZKcLnX.png
The modded atiflash that guys did also works as intended. Tweaking the timings did increase Superposition and 3DMark Timespy scores for me slightly too.
I need to test out the 1250Mhz straps on the weekend and try to decode the straps somehow to test them one by one.

[PIOUPIOU99]

Dear hulk999, thank you for your input. I can confirm the Gigabyte RX Vega 64 Gaming OC is problematic for Ethereum mining in particular. The memory configuration (mine is also Samsung), seems to refuse overclocking or higher performance straps. The best I could achieve was 990mhz on the memory using stock straps. In other words just over 38 mega hashes, like you. Yes, I have upgraded my bios from version F1 to F2.

Thank you for referring to team red miner and their vega timing options. I now for the first time am achieving 2200 hash mining Monero. Although I would prefer to mine Ethereum at 50 mega hashes!

So, if anyone in this thread has any clue why some of us with Gigabyte Gaming OC Vega's have trouble with Ethereum tuning please respond. Good day to all.

if you are able to achieve 2.2kh on CN/R there's absolutely no reason not to be able to run ETH with over 45MH... CNR should be more demanding on the memory in terms of stability so likely you are doing something wrong
1) apply ppt that will set default clocks and volts much lower
2) for fine tuning of volts and frequencies you may use overdriven tool. core should be around 800-850mv, freq not too high, over 1300MHz is likely overkill for ETH, mem as high as you can get it stable, SOC should be @1.1GHz
3) after it's stable around 42MH for vega64, with the settings above, you can try amdmemtweak and go to ~50

EDIT: and do not use Afterburner - have it disabled from Task Scheduler (also do DDUs and have clean driver setups, preferably without the AMD settings GUI crap)

I have exactly the same concerns on this card, I do not understand how it works. My msi 56 and sapphire 64 are perfect under eth with trm but there I am lost.
no powerplay table works properly, blue screen or frozen win. I cannot exceed 950/970 in memory. first time for me .possible to push at 40mh but after no tweak passes without crushing of trm or win .10.
I don't understand this card

edit :gpu dead .

[Eliovp]

So.. I guess we'll see 5700 bios modding in the future
https://www.overclock.net/forum/28308884-post445.html
https://amia.io/2020/01/03/rx5700-mod-bios-test-eth-mining/
http://dr-shost.com/files/atiflash293plus.zip
http://dr-shost.com/files/rx5700-orig-mod-bios.zip

No you won't..

Flashing an XT vbios to a non XT card is the only thing that is possible.

The onboard PSP prevents you from doing modifications.

Also, memory timing changes on a Navi are like on an R7, useless.

The only thing that would improve performance (particularly eth mining wise) would be to make it so that you extend your bandwith.
Which means that you would need to change clocks + timings so it remains stable and because the PSP on board even has it's own PPtable, this is yet again not possible.

Well, anything is possible, or almost :p
But at this stage, i doubt anyone has been able to accomplish vbios edits for navi (basically bypass PSP).

Cheers

[ctsbitcoin]

Ugh.  Been a TUBE/CCX miner and trying to get ETH to be efficient.  I got 4x V64 but all I get is 36MHs each where I've read it is possible to get 50MHs+.  

https://git.mmpos.eu/windows/teamredminer/blob/master/ETHASH_GENERAL_TUNING.txt

Upon finding the AMD Memory Tweak XL, I use:

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

I get a crash.  I've tried settings individually, no change.

P7 1075 / 850 as suggested or 1408 / 925 shows no change.

P3 1107 / 850 as suggested crashes whereas 945 / 1050 appears to be fine but stuck at 36MHs.

Embarrassingly, my Red Dragon V56s get 39MHs easily without any but P3/P7 and at -50%!

My question is how do I get my V64s tweaked correctly?