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

[boxraider]

Is there any solution for the non ref 56 cards that have a default of 800 mem and can’t go past 900-950?


I got four of these crap cards any thoughts ? Three are red devil 56 power colors and one is a sapphire pulse

Memory tweak isn't exactly working for me so this isn't fully optimized yet as I can only experiment in the rare cases when it happens to work after a reboot. However, the following seems to be working great for me with such cards (assuming samsung mem).

Doesn't affect stability at all, no errors, adds 200 h/s give or take per card on XMR. It is just a variation of the stock. I hope it helps.

Code:

--RAS 20 --RCDRD 13 --RCDWR 6  --RC 37 --RP 12 --RRDS 4 --RRDL 5 --RFC 208 --REF 15600

im having the same issue. I had my timings changed just to the preset level 1 but now its at default. it'll just say timing level 1 or 2 depending on what i change it to but it wont show up on the timing inputs on the left side. Still curious on how to address this problem, been using wattman for timing preset changes and it will do the exact same thing but not change my timings when i pull up AMDTweak

[1623574848]

1623574848

[1623574848]

1623574848

[1623574848]

1623574848

[Eliovp]

Is there any solution for the non ref 56 cards that have a default of 800 mem and can’t go past 900-950?

I got four of these crap cards any thoughts ? Three are red devil 56 power colors and one is a sapphire pulse

If you're lucky, you might find a Vega 64 vbios for that same card (if they made em both 56 & 64), which is unique.. not much out there, Pulse, only 56 afaik.

That however doesn't mean they're crappy, i've got a Sapphire Pulse with Hynix memory and it's running smooth at 47~Mh (memory at 975).
Got a Samsung Vega56 too and hits 48+Mh..

The thing is, they only have 6 SOC states, so it gets auto forced into a lower memory state (800MH), you can see this in the tool..
Vega 64 has 8 SOC states, therefor it goes to the highest memory state (1000Mh).

is there way to apply straps like the claymore miner ?
for polaris

GUI version has an export/import feature, you can use that.. "kind" of the same.

Hello,

I have strange problem with one rig with 3 vega 64 liqued cooling.
I try many memory tweaks with latest memtweak cli and got 51,5 Mh/s on each vega, but pools (ethermine, 2miners, miningpoolhub) show disconected status. But if I lucky and pool accept connections -> low end pool hashrate
HBM2 1100 and CORE 1050 on all three cards (Win10x64 1809). Tried claymore and phoenix. Coins - etc, eth, exp
When i reset timings - all card hashing without problems on pool.

Thank you in advance for any advice.

I'm not able to reproduce this, just to be sure i let my Vega 64 run at 53Mh for half an hour, pool hashrate was spot on.

What are your timings?

Cheers!

[Marvell2]

@Eliovp "magic trick" is to disable p6 and p7 core state and stay on p5 core , p6 and p7 jump one soc clock step higher on some men frequency up to (1107soc) (example p7 core clock and p3 mem and your frequency is 830 hbm mem so your step is higher then 800soc so 847soc +"1 soc step" so you will get 900soc if you set p5 core you will get 847soc )

please explain this again , so yo usay on a 56 disable the p6 and p7 states on core then i can set higer memory than the 900 i seem to be capped at ? or you are saying by disabling those two the mem will automatically jump higer ?

**edit*** @duck Hellen
You are definitely on to something, I disabled the p6 an p7 states for my 4 card Hynix 56 rig and all of a sudden no more crashes or instability, power draw is also good 150 watts from the wall, I’m still at 1100/825mv and 900/825

Hash rate is 42 MHz on eth a bit higer than before.  Timing are all the same only thing I did was disable p6 an p7

Can I attempt to bump up the mem now ?

[rednoW]

@Eliovp "magic trick" is to disable p6 and p7 core state and stay on p5 core , p6 and p7 jump one soc clock step higher on some men frequency up to (1107soc) (example p7 core clock and p3 mem and your frequency is 830 hbm mem so your step is higher then 800soc so 847soc +"1 soc step" so you will get 900soc if you set p5 core you will get 847soc )

please explain this again , so yo usay on a 56 disable the p6 and p7 states on core then i can set higer memory than the 900 i seem to be capped at ? or you are saying by disabling those two the mem will automatically jump higer ?

**edit*** @duck Hellen
You are definitely on to something, I disabled the p6 an p7 states for my 4 card Hynix 56 rig and all of a sudden no more crashes or instability, power draw is also good 150 watts from the wall, I’m still at 1100/825mv and 900/825

Hash rate is 42 MHz on eth a bit higer than before.  Timing are all the same only thing I did was disable p6 an p7

Can I attempt to bump up the mem now ?

940 mem clock is max I was able to get stable with hynix and std v56 1.25 mem voltage

[rednoW]

If you're lucky, you might find a Vega 64 vbios for that same card (if they made em both 56 & 64), which is unique.. not much out there, Pulse, only 56 afaik.

That however doesn't mean they're crappy, i've got a Sapphire Pulse with Hynix memory and it's running smooth at 47~Mh (memory at 975).
Got a Samsung Vega56 too and hits 48+Mh..

I have my Sapphire Pulse Vega56 Hynix one bios flashed to Nitro 64 for eth, it is working ok at 1040 mem clock than.
But for cryptonight it is better to stay with original bios for low power with almost the same hash rate.

[lstrike]

I have my Sapphire Pulse Vega56 Hynix one bios flashed to Nitro 64 for eth, it is working ok at 1040 mem clock than.
But for cryptonight it is better to stay with original bios for low power with almost the same hash rate.

So you put samsung bios on Hynix card?

[Miner Addict]

Is there any solution for the non ref 56 cards that have a default of 800 mem and can’t go past 900-950?

I got four of these crap cards any thoughts ? Three are red devil 56 power colors and one is a sapphire pulse

If you're lucky, you might find a Vega 64 vbios for that same card (if they made em both 56 & 64), which is unique.. not much out there, Pulse, only 56 afaik.

That however doesn't mean they're crappy, i've got a Sapphire Pulse with Hynix memory and it's running smooth at 47~Mh (memory at 975).
Got a Samsung Vega56 too and hits 48+Mh..

The thing is, they only have 6 SOC states, so it gets auto forced into a lower memory state (800MH), you can see this in the tool..
Vega 64 has 8 SOC states, therefor it goes to the highest memory state (1000Mh).

is there way to apply straps like the claymore miner ?
for polaris

GUI version has an export/import feature, you can use that.. "kind" of the same.


Wow...975 mclock, Wattage ?
Soft PP or 64 bios ?

Btw Wonderful

Hello,

I have strange problem with one rig with 3 vega 64 liqued cooling.
I try many memory tweaks with latest memtweak cli and got 51,5 Mh/s on each vega, but pools (ethermine, 2miners, miningpoolhub) show disconected status. But if I lucky and pool accept connections -> low end pool hashrate
HBM2 1100 and CORE 1050 on all three cards (Win10x64 1809). Tried claymore and phoenix. Coins - etc, eth, exp
When i reset timings - all card hashing without problems on pool.

Thank you in advance for any advice.

I'm not able to reproduce this, just to be sure i let my Vega 64 run at 53Mh for half an hour, pool hashrate was spot on.

What are your timings?

Cheers!

[ganzocrypt]

Hello,
I played with tweaking core and memory for RX580 8Gb and 4Gb with hynex mem, but cannot reach an optimal setup with timing.

Isn't possible to do like claymore is doing?

Recognizing the GPU and Memory and than applying a "optimal" tweak ?
If there are enough data from the community for the various cards, would that be possible?

thx
great tool btw !

[naska21]

Thank you both for the suggestions.

Zorg33, I checked the files and they seemed alright (date, size, everything). I still went ahead and copied the file from C:\AMD just to make sure, still no change I'm afraid. I also uninstalled everything w/DDU, double-checked the files were deleted and installed the driver again. Again, no change.

naska21, could you elaborate a bit? The system is running fine, mining for days non-stop, I just can't make AMD Memory Tweak work for whatever reason. I could be missing something of course.

edit: forgot to mention, no conflicts in device manager.

Oh, I thought you have faced  the spontaneous loss of working cards in your rig but if they all  are non-stop running then everything is OK  with IRQ sharing. Sorry for misunderstanding. As to memory tweaks, I've never known such a situation. Arguably you have quite a bunch of cards attached to motherboard.

[boxraider]

Hello,
I played with tweaking core and memory for RX580 8Gb and 4Gb with hynex mem, but cannot reach an optimal setup with timing.

Isn't possible to do like claymore is doing?

Recognizing the GPU and Memory and than applying a "optimal" tweak ?
If there are enough data from the community for the various cards, would that be possible?

thx
great tool btw !

Hey Ganzo, i did some research and hynix is slower than micron and samsung cards. apparently its a toss-up too, you can get 2 MSI brand graphics card of exact model and make and it could end up a hynix and samsung mix. to me that just means economics at the foundries. sucks for us!  im getting 25mhs with underclock and undervolt. i barely overclocked mine, just in case hynix cant operate at a higher frequency than samsung/micron cards. CoreClock 1146mhz/860mv. MemClock 2050, pulling 109w + 16w. Could anyone tell me that this is as good as i can get it given that its a hynix?

But i am planning on buying a radeon vii and was wondering since its a Vega II, would the Vega strap work on this tool?? Thanks guys!

[SamAlackass]

Thank you both for the suggestions.

Zorg33, I checked the files and they seemed alright (date, size, everything). I still went ahead and copied the file from C:\AMD just to make sure, still no change I'm afraid. I also uninstalled everything w/DDU, double-checked the files were deleted and installed the driver again. Again, no change.

naska21, could you elaborate a bit? The system is running fine, mining for days non-stop, I just can't make AMD Memory Tweak work for whatever reason. I could be missing something of course.

edit: forgot to mention, no conflicts in device manager.

Oh, I thought you have faced  the spontaneous loss of working cards in your rig but if they all  are non-stop running then everything is OK  with IRQ sharing. Sorry for misunderstanding. As to memory tweaks, I've never known such a situation. Arguably you have quite a bunch of cards attached to motherboard.

No worries, thank you for the input regardless. It's a weird issue really, and way too random to try and make any sense of it. One day I think it must be hardware related, next day I wonder how can it work when it does. It's only four cards BTW, maybe I still need to try and remove them one by one when I get a chance, just to make sure. An option to ignore the onboard GPU would be nice too.. Eliovp

[seefatlow]

1150Mhz core at 1150mV, isnt that too much? whats is best value? im using 850mV and when not stable lowering to 1120Mhz. whats it proper voltage for 1200Mhz core?

I'm very interrested in lowering power consumption for 570 and 580s. On my nvidias i'm able to get 270Mh/s at 750w (would be lower, but cooling fans makes about 80-100w!)
On my RX rigs it is trrible, I get ~1650w at 350Mh/s while in cool environment it can get as low as 1450w so 200w just for cooling!!, but still 1450 base for one 12gpu rig seems very high to me. Also in this case I have 3 psu from times of dual mining, and only 80plus gold, so that can be quite a lot wasted, maybe another 200w? I don't have physical access to switch to 2 psus anymore and even if I did there is risk of some gpus not booting and time to troubleshoot it just not worth it.

I'm writing about real power consumtion, aproximated from about 24h. Not some virtual numbers from hiveOS/hwinfo etc. My numbers includes about 20-25w of "system" consumption.

I'm not able to get 850mV@1150mhz at any of my amd cars, i think 880mV (I tried lowering mhz to 1100 and gpus still crash, but upping voltage helps) is lowest i'm able to run stable. Stable for me means at least 15days without miner restart, i don't measure stability in hours.


So as you can see, with my rate of about 0.85usd/kw I barely manage to be profitable. Still easier then selling rigs though.

I am interested to know how you are able to pull 270MH/s at just 750W !!!! What cards are you using? Please share.

[Marvell2]

Anyone have timings for frontier Vegas? They use different ram than the 64s and the bottom number is 350 not 260 like the Samsung vegas.

Most I can get by using the same settings is 46 MHz but stock with power play settings 1100 mem  they get 42 so it’s not much of an improvement at all

[rednoW]

Anyone have timings for frontier Vegas? They use different ram than the 64s and the bottom number is 350 not 260 like the Samsung vegas.

Most I can get by using the same settings is 46 MHz but stock with power play settings 1100 mem  they get 42 so it’s not much of an improvement at all

Mine cryptonight-heavy algos with them. No need for dagger

[ganzocrypt]

Hello,
I played with tweaking core and memory for RX580 8Gb and 4Gb with hynex mem, but cannot reach an optimal setup with timing.

Isn't possible to do like claymore is doing?

Recognizing the GPU and Memory and than applying a "optimal" tweak ?
If there are enough data from the community for the various cards, would that be possible?

thx
great tool btw !

Hey Ganzo, i did some research and hynix is slower than micron and samsung cards. apparently its a toss-up too, you can get 2 MSI brand graphics card of exact model and make and it could end up a hynix and samsung mix. to me that just means economics at the foundries. sucks for us!  im getting 25mhs with underclock and undervolt. i barely overclocked mine, just in case hynix cant operate at a higher frequency than samsung/micron cards. CoreClock 1146mhz/860mv. MemClock 2050, pulling 109w + 16w. Could anyone tell me that this is as good as i can get it given that its a hynix?

But i am planning on buying a radeon vii and was wondering since its a Vega II, would the Vega strap work on this tool?? Thanks guys!

I know Hynex is not the best but I got to 31 Mh/s, so wanted to see if the AMD Tweak tool increase that.

It would be good if the tool does what Claymore has, like a DB for common cards, with some "optimal" / "baseline" strap and than one can tweak it some more based on OC and cards differences, for example silicon!

Hopefully we get there with the tool

[Geezartheminer]

Hello all,

I noted in an earlier post issue running the CLI via a batch file, sadly I have the same issue - has anyone come across this (seemingly basic Windows/Authentication issue) before? in that post the admin rights were applied to a batch file that would then run the batch file with correct access privs.

Batch File Usage Issue
Problem - unable to launch the CLI via batch file via a simple batch file in the start-up folder (current up to date patched Win 10 build)
Workaround - manually launching the batch file works, settings applied and CLI/GUI work as expected. I am able to use CMD to edit/tweak/test and run the batch file manually but not the start-up - all other miners/use of start-up has never failed until now

Straps Issue
Straps when viewed from a Polaris BIOS are displayed correctly/setup and strap working correctly it would seem but less 200 H/s  GUI displays the default settings (even under load/running the miner) stumped by this as when I manually re-apply the same settings on the strap and re-run the CLI/GUI tool the correct settings are applied (over the BIOS strap defaults) - very strange - strap when viewed in R_ encoded and re-added - again same issue. (straps on the bios have the edited tweaks but are not working unless the GUI tool/CLI tool re-adds them - stumped lol

So GUI not reporting the settings as they are on the strap on the modded BIOS (just the default) not on all Polaris, but on a mining 470 (tried all 4 BIOS switches - same) otherwise when I edit the strap, apply - they work / as expected


Thanks all

[MegaEvgeniy]

Hello,

I have strange problem with one rig with 3 vega 64 liqued cooling.
I try many memory tweaks with latest memtweak cli and got 51,5 Mh/s on each vega, but pools (ethermine, 2miners, miningpoolhub) show disconected status. But if I lucky and pool accept connections -> low end pool hashrate
HBM2 1100 and CORE 1050 on all three cards (Win10x64 1809). Tried claymore and phoenix. Coins - etc, eth, exp
When i reset timings - all card hashing without problems on pool.

Thank you in advance for any advice.

I'm not able to reproduce this, just to be sure i let my Vega 64 run at 53Mh for half an hour, pool hashrate was spot on.

What are your timings?

Cheers!

Hello, thank you for answer. Found problem - to many incorects shares. Too high mem overclock.

[SamAlackass]

Hello all,

I noted in an earlier post issue running the CLI via a batch file, sadly I have the same issue - has anyone come across this (seemingly basic Windows/Authentication issue) before? in that post the admin rights were applied to a batch file that would then run the batch file with correct access privs.

Batch File Usage Issue
Problem - unable to launch the CLI via batch file via a simple batch file in the start-up folder (current up to date patched Win 10 build)
Workaround - manually launching the batch file works, settings applied and CLI/GUI work as expected. I am able to use CMD to edit/tweak/test and run the batch file manually but not the start-up - all other miners/use of start-up has never failed until now

Straps Issue
Straps when viewed from a Polaris BIOS are displayed correctly/setup and strap working correctly it would seem but less 200 H/s  GUI displays the default settings (even under load/running the miner) stumped by this as when I manually re-apply the same settings on the strap and re-run the CLI/GUI tool the correct settings are applied (over the BIOS strap defaults) - very strange - strap when viewed in R_ encoded and re-added - again same issue. (straps on the bios have the edited tweaks but are not working unless the GUI tool/CLI tool re-adds them - stumped lol

So GUI not reporting the settings as they are on the strap on the modded BIOS (just the default) not on all Polaris, but on a mining 470 (tried all 4 BIOS switches - same) otherwise when I edit the strap, apply - they work / as expected


Thanks all

For your first issue, right click the .exe, go to properties, compatibility, tick the "always run as admin". You still need to disable UAC I guess, alternatively you could use task scheduler to run as admin at logon or each boot.

Can't help much with your second issue I'm afraid, hopefully someone else will though.

[Geezartheminer]

Thanks will have a look at why its not working again today - 1st time came across such an issue - otherwise very pleased with the tool

[boxraider]

Hello,
I played with tweaking core and memory for RX580 8Gb and 4Gb with hynex mem, but cannot reach an optimal setup with timing.

Isn't possible to do like claymore is doing?

Recognizing the GPU and Memory and than applying a "optimal" tweak ?
If there are enough data from the community for the various cards, would that be possible?

thx
great tool btw !

Hey Ganzo, i did some research and hynix is slower than micron and samsung cards. apparently its a toss-up too, you can get 2 MSI brand graphics card of exact model and make and it could end up a hynix and samsung mix. to me that just means economics at the foundries. sucks for us!   im getting 25mhs with underclock and undervolt. i barely overclocked mine, just in case hynix cant operate at a higher frequency than samsung/micron cards. CoreClock 1146mhz/860mv. MemClock 2050, pulling 109w + 16w. Could anyone tell me that this is as good as i can get it given that its a hynix?

But i am planning on buying a radeon vii and was wondering since its a Vega II, would the Vega strap work on this tool?? Thanks guys!

I know Hynex is not the best but I got to 31 Mh/s, so wanted to see if the AMD Tweak tool increase that.

It would be good if the tool does what Claymore has, like a DB for common cards, with some "optimal" / "baseline" strap and than one can tweak it some more based on OC and cards differences, for example silicon!

Hopefully we get there with the tool

Hey ganzo, I had 31mhs going for a minute and tried a small 5mv undervolt and it crashed. I'm not able to get back to the same hash rate as before.. and was wondering how you got yours up to 31mhs so i can replicate it? I had my orders of operation down on paper but it just seemed to be fussy with me.  

But, does anyone know what i am doing wrong when im trying to change my tRFC settings from base 219 to any setting and itll just revert back to 219. I've check my HWinfo and it says my tRFC is at 467T. I'm trying to bring it down but its way up. I am using the win gui version and am a little retarded, and dont know how to use the cli version. (pardon my french)