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

[pbfarmer]

Hello, is anyone having an issue where the last GPU of a multi-GPU rig does not get the mem tweak applied? On 3 out of my 4 rigs (6 Vega64s per rig), the last GPU does not get the tweak applied to it. I’m using CLI (last version) on Win10.

Double-check you enumerate your GPUs correctly. I know it sounds like a silly suggestion, but you know... been there, done that.

If the onboard GPU is anything AMD, it gets listed too (usually GPU0).  

I hear ya, made plenty of stupid mistakes myself. But yeah, TRM miner shows the bus number and changes I make to the other GPUs do take effect. I’m wondering if it’s because I’m using CLI and the - -i parameter and it’s not reading the last GPU in the parameter setting?

You don't happen to start TRM with a -d in config?

No, but I’m going to do some further tests tomorrow with the latest version and see what happens, maybe double check the settings with the GUI (I’m using CLI right now).

You're using '--i' and not '-i', correct?  The example usage in the help text (amdmemtweak --help) shows the latter, which is not valid.  (EDIT: this was the case in older versions - appears to be fixed in the latest)

[1713984042]

1713984042

[1713984042]

1713984042

[livada]

vega samsung memory : ETH-CNR-BEAM

WinAMDTweak.exe --RAS 28 --RCDRD 12 --RCDWR 5 --RC 44 --RP 12 --RRDS 3 --RRDL 3 --REF 15600 --RFC 246 --FAW 14

http://imgbox.com/oH0Nn1Pa

Hi livada, thanks for all your posts and suggestions, I’m now mining 20-25% more hash at the pool because of you (and of course Eliovp) :-)

But when you run settings tests, you need to run them longer. 1 accepted share does not make a for a very good basis for overall long-term results. Valid results would be based on hours of mining, sometimes GPU crashes with new settings after a while, sometimes many hw errors occur over time, these may not present themselves until many hours of mining.

Cheers!

I test this  1-2 day. This is only SS for send..
With this amd mem tweak setup my 3*vega56 powercolor and 1*vega64LE (all samsung mem) work fine with many algo- I test  CRN-GRAFT- TRTL- MOX-WOW-MSR-LITEv7. any algo have better hr if i change mem tweak setup but with this  setings all algo work fine.

I have a Win 10 - Vega 56 rig

One Hynix card causes a dead GPU when trying 19.4.2, latest TR and have tried both of the suggested Hynix setttings.  Anyone have one that is working for them?   I've tried:

WinAMDTweak.exe --gpu 03 --RAS 24 --RCDRD 19 --RCDWR 4 --RC 35 --RP 13 --RRDS 4 --RRDL 5 --RFC 148 --REF 15600

and

WinAMDTweak.exe --gpu 03 --RAS 22 --RCDRD 17 --RCDWR 4 --RC 35 --RP 13 --RRDS 4 --RRDL 4 --RFC 148 --REF 15600

Both of these cause a dead gpu/restart

try this-

--rcdrd 19 --rcdwr 4 --rc 35 --rp 14 --rrds 4 --rrdl 5 --rfc 148 --REF 15600
or
--rp 14 --rc 42 --rfc 156 --rrd 3 --rcdrd 18 --rcdwr 3

not best but work

iff any have good setup for vega56/stockbios/hynix pls send ..

I got vega 56 strix hynix and i cant go more than 1800hs.. What is ur best results guys? Any good setup for this bad cards?

my 6*veg56 nitro+(hynix) work fine with this setings (+2000hr)- 1 card give 1 error every 1 hour but work fine. itest this with more algo.

WinAMDTweak.exe --rp 14 --rc 35 --rfc 178 --rrds 4 --rrdl 5 --rcdrd 19 --rcdwr 5 --REF 15600

if not work try add --RAS 44(gives stability and lower HR) or chg --rfc 178 to --rfc 188(lower HR)

best card  have gpu= 1440mhz/866mv mem=935mhz/868mv
bad card have   gpu= 1400mhz/870mv mem=895mhz/875mv and give error every 1hour
http://imgbox.com/4Ast1m6p

[Ravenoss]

vega samsung memory : ETH-CNR-BEAM

WinAMDTweak.exe --RAS 28 --RCDRD 12 --RCDWR 5 --RC 44 --RP 12 --RRDS 3 --RRDL 3 --REF 15600 --RFC 246 --FAW 14

http://imgbox.com/oH0Nn1Pa

Hi livada, thanks for all your posts and suggestions, I’m now mining 20-25% more hash at the pool because of you (and of course Eliovp) :-)

But when you run settings tests, you need to run them longer. 1 accepted share does not make a for a very good basis for overall long-term results. Valid results would be based on hours of mining, sometimes GPU crashes with new settings after a while, sometimes many hw errors occur over time, these may not present themselves until many hours of mining.

Cheers!

I test this  1-2 day. This is only SS for send..
With this amd mem tweak setup my 3*vega56 powercolor and 1*vega64LE (all samsung mem) work fine with many algo- I test  CRN-GRAFT- TRTL- MOX-WOW-MSR-LITEv7. any algo have better hr if i change mem tweak setup but with this  setings all algo work fine.

I have a Win 10 - Vega 56 rig

One Hynix card causes a dead GPU when trying 19.4.2, latest TR and have tried both of the suggested Hynix setttings.  Anyone have one that is working for them?   I've tried:

WinAMDTweak.exe --gpu 03 --RAS 24 --RCDRD 19 --RCDWR 4 --RC 35 --RP 13 --RRDS 4 --RRDL 5 --RFC 148 --REF 15600

and

WinAMDTweak.exe --gpu 03 --RAS 22 --RCDRD 17 --RCDWR 4 --RC 35 --RP 13 --RRDS 4 --RRDL 4 --RFC 148 --REF 15600

Both of these cause a dead gpu/restart

try this-

--rcdrd 19 --rcdwr 4 --rc 35 --rp 14 --rrds 4 --rrdl 5 --rfc 148 --REF 15600
or
--rp 14 --rc 42 --rfc 156 --rrd 3 --rcdrd 18 --rcdwr 3

not best but work

iff any have good setup for vega56/stockbios/hynix pls send ..

I got vega 56 strix hynix and i cant go more than 1800hs.. What is ur best results guys? Any good setup for this bad cards?

my 6*veg56 nitro+(hynix) work fine with this setings (+2000hr)- 1 card give 1 error every 1 hour but work fine. itest this with more algo.

WinAMDTweak.exe --rp 14 --rc 35 --rfc 178 --rrds 4 --rrdl 5 --rcdrd 19 --rcdwr 5 --REF 15600

if not work try add --RAS 44(gives stability and lower HR) or chg --rfc 178 to --rfc 188(lower HR)

best card  have gpu= 1440mhz/866mv mem=935mhz/868mv
bad card have   gpu= 1400mhz/870mv mem=895mhz/875mv and give error every 1hour
http://imgbox.com/4Ast1m6p

What miner u use? I use last release of Srbminer and results very bad 

[livada]

What miner u use? I use last release of Srbminer and results very bad  

TRM for all suport(v8), phoenix for eth, gminer for beam and srb for other  alfgo like litev7-mox-wow

[Iamtutut]

New Release

• Added FIJI support
• Added Refresh Rate timing (RX boost  )
• Overdrive fixes
• Added memory manufacturer detection for G5, HBM & HBM2
• .....

Link: Here


Will upload CLI versions asap as well!


I noticed some people have figured out that there is still room to improve on RX 4**/5** cards.
Just not many people saying how..

Well, open up the tool, and raise Refresh Rate (REF) i created a separate textbox for it to make it easy.
Raise it from 5 to 7/8 (just an example) and you'll see improvement already.

Slightly dropping RFC will increase your hashrate as well.


Greetings

It's slightly faster, thanks a lot !
I applied my usual settings for RX574.

tRFC:
- Samsung memory (tCL is 22): 117
- Hynix memory (tCL is 19): 124
- Elpida memory (tCL is 21): 94

Right now refresh at 9 for all GPUs. I'll see if it's stable.

These settings rae very stable to me, rehresf at 8 for Samsung memory, 9 for Hynix and Elpida.

Eliovp, do you accept tips in TRTL (turtlecoin) ?

[jimmyD30]

Hello, is anyone having an issue where the last GPU of a multi-GPU rig does not get the mem tweak applied? On 3 out of my 4 rigs (6 Vega64s per rig), the last GPU does not get the tweak applied to it. I’m using CLI (last version) on Win10.

Double-check you enumerate your GPUs correctly. I know it sounds like a silly suggestion, but you know... been there, done that.

If the onboard GPU is anything AMD, it gets listed too (usually GPU0).  

I hear ya, made plenty of stupid mistakes myself. But yeah, TRM miner shows the bus number and changes I make to the other GPUs do take effect. I’m wondering if it’s because I’m using CLI and the - -i parameter and it’s not reading the last GPU in the parameter setting?

You don't happen to start TRM with a -d in config?

No, but I’m going to do some further tests tomorrow with the latest version and see what happens, maybe double check the settings with the GUI (I’m using CLI right now).

You're using '--i' and not '-i', correct?  The example usage in the help text (amdmemtweak --help) shows the latter, which is not valid.  (EDIT: this was the case in older versions - appears to be fixed in the latest)

Thank you for the suggestion, yes I’m using the double dash.

[jimmyD30]

I test this  1-2 day. This is only SS for send..
With this amd mem tweak setup my 3*vega56 powercolor and 1*vega64LE (all samsung mem) work fine with many algo- I test  CRN-GRAFT- TRTL- MOX-WOW-MSR-LITEv7. any algo have better hr if i change mem tweak setup but with this  setings all algo work fine.

Thank you for the clarification :-)

[Ravenoss]

What miner u use? I use last release of Srbminer and results very bad  

TRM for all suport(v8), phoenix for eth, gminer for beam and srb for other  alfgo like litev7-mox-wow

Thank you very much, i got 2000hs for card, very good miner and timings

[zeo229]

I can't run the program (gui). Get an excetpion:
StartService failed: The service cannot be started,
either because it is disabled or because it has no enabled devices associated with it
at WinFormsGUI.IOMap.Loaddriver()
at WinFormsGUI.Program.Main()



What could be the issue?

[rednoW]

Is it only for me that loading profile with latest V5 results in fan speeds blank table at the right bottom ?
Polaris cards. V4 was and is OK.

[pbfarmer]

New Release

• Added FIJI support
• Added Refresh Rate timing (RX boost  )
• Overdrive fixes
• Added memory manufacturer detection for G5, HBM & HBM2
• .....

Link: Here


Will upload CLI versions asap as well!


I noticed some people have figured out that there is still room to improve on RX 4**/5** cards.
Just not many people saying how..

Well, open up the tool, and raise Refresh Rate (REF) i created a separate textbox for it to make it easy.
Raise it from 5 to 7/8 (just an example) and you'll see improvement already.

Slightly dropping RFC will increase your hashrate as well.


Greetings

It's slightly faster, thanks a lot !
I applied my usual settings for RX574.

tRFC:
- Samsung memory (tCL is 22): 117
- Hynix memory (tCL is 19): 124
- Elpida memory (tCL is 21): 94

Right now refresh at 9 for all GPUs. I'll see if it's stable.

These settings rae very stable to me, rehresf at 8 for Samsung memory, 9 for Hynix and Elpida.

Eliovp, do you accept tips in TRTL (turtlecoin) ?

I’m currently benching  REF 10 across all my 580s (samsung, micron, and hynix) - 3 errors out of 1496 results total (.2%) so far.  They are actually from 2 cards specifically - a Samsung (2/192) and a Hynix (1/171).

Gain looks to be ~0.7-0.8Mh/s (ethash) per card, or about 2-2.5%

[tale01]

What config for Fiji ?  

I follow picture on github, it cannot works.

[spamster]

I can't run the program (gui). Get an excetpion:
StartService failed: The service cannot be started,
either because it is disabled or because it has no enabled devices associated with it
at WinFormsGUI.IOMap.Loaddriver()
at WinFormsGUI.Program.Main()



What could be the issue?

Zeo, I have the same problem on one machine I am trying to diagnose.  I have 2 identical machines, it works on one and I get that error on the other.  I have DDU'ed and reinstalled drivers on both machines.  Both machines have enabled onboard Intel graphics, disabling has no effect.  I will post more as I discover.

[pbfarmer]

@Eliovp - a couple bugs/issues to report (latest version, linux, gddr5):

1. setting --RAS reports as having set --PA2RDATA.  Looks like maybe just a display issue tho, as --current then reports RAS as having been set properly

2. setting --RC is also applying the value to CL - and since setting CL doesn't work (prevents card from being initialized by miner, reboot required) effectively RC cannot be set

3. you might be missing some bits for RC (or it's just incorrectly mapped to CL)?  strap says 65, but it reads same values as CL, and if i set > 31, i get an overflow (e.g. setting 32 results in 0).  

EDIT: Btw - any hints why DRAM1/2 values change under load, even when the core/mem state haven't changed?

[lstrike]

Gain looks to be ~0.7-0.8Mh/s (ethash) per card, or about 2-2.5%

(Polaris)
What is the power increase for that?
The Claymore boost increases exponentially the power consumption for similar extra hash, which is not effective...

[boxraider]

Hey, guys

Any tips on how to get the tool to make the timing settings stick? What programs are you guys running incl. this tool? I've got afterburner, and amd tweak. It's weird, I have to set my core clock in after burner and then tweak my voltages, memory and one-click timings in amd tweaker. But, 80% of the time itll crash or not apply one-click timings/manual timing changes when im running the same order of operations (weird 20%ish times my one-click will stick and reflect on the timing chart). It seems that I might have a bad card? Im not sure how you guys are able to adjust your timings and having them show up on your charts.

Right now, Im just getting 31mhs rather than crashing my pc 40x and settling for 28mhs (no 1 click timing/manual timing change) because Ive given up for the day lol.

[Eliovp]

@Eliovp - a couple bugs/issues to report (latest version, linux, gddr5):

1. setting --RAS reports as having set --PA2RDATA.  Looks like maybe just a display issue tho, as --current then reports RAS as having been set properly

2. setting --RC is also applying the value to CL - and since setting CL doesn't work (prevents card from being initialized by miner, reboot required) effectively RC cannot be set

3. you might be missing some bits for RC (or it's just incorrectly mapped to CL)?  strap says 65, but it reads same values as CL, and if i set > 31, i get an overflow (e.g. setting 32 results in 0).  

EDIT: Btw - any hints why DRAM1/2 values change under load, even when the core/mem state haven't changed?

I'm going to check this tomorrow, did everything in a rush without even testing  

Should clean a bit up because it's becoming a mess.
Already saw some issues going over the code.

And Laik2 pointed out that some bits are wrong, he's right (PMG for example.. 3-1-3-1-4-6-3..)
(See that Laik2, i'm not too "Godlike" to acknowledge my mistakes)

I'll see to add some extra timings that i didn't yet expose, will expose HBM2 Memory Manufacturer as well.

Anyhow, when i wake up i'll take some more time to test and clean up code because after adding hbm1 it's a total mess.

Cheers!

[pbfarmer]

Gain looks to be ~0.7-0.8Mh/s (ethash) per card, or about 2-2.5%

(Polaris)
What is the power increase for that?
The Claymore boost increases exponentially the power consumption for similar extra hash, which is not effective...

Nothing - basically increasing REF should just be replacing refresh reads/writes with workload reads/writes, so you really shouldn't see any power use change, unless your RFC was so high that you have a bunch of NOOPs during the refresh period.

FYI, my power use is ~130w for 32.25Mh/s per card avg (which seems a bit high to me, given that i'm only at 1175 cclock, 2100-2200 mclock, and 837mv avg, tho it's a little warmer today, so I think my fans are pulling a bit more)

[pbfarmer]

Hey, guys

Any tips on how to get the tool to make the timing settings stick? What programs are you guys running incl. this tool? I've got afterburner, and amd tweak. It's weird, I have to set my core clock in after burner and then tweak my voltages, memory and one-click timings in amd tweaker. But, 80% of the time itll crash or not apply one-click timings/manual timing changes when im running the same order of operations (weird 20%ish times my one-click will stick and reflect on the timing chart). It seems that I might have a bad card? Im not sure how you guys are able to adjust your timings and having them show up on your charts.

Right now, Im just getting 31mhs rather than crashing my pc 40x and settling for 28mhs (no 1 click timing/manual timing change) because Ive given up for the day lol.

I load PPTs to set clocks, lock to the power states I use while mining (i use cclock 6 and mclock 2 for ETH on polaris), then apply timings w/ amdmemtweak (cli).  Changing to a different state (core or mem) at any point seems to lose the applied timings.

I'm on linux, but I would expect things to be basically the same in Windows...

[seefatlow]

vega samsung memory : ETH-CNR-BEAM

WinAMDTweak.exe --RAS 28 --RCDRD 12 --RCDWR 5 --RC 44 --RP 12 --RRDS 3 --RRDL 3 --REF 15600 --RFC 246 --FAW 14

Hi livada, thanks for all your posts and suggestions, I’m now mining 20-25% more hash at the pool because of you (and of course Eliovp) :-)

But when you run settings tests, you need to run them longer. 1 accepted share does not make a for a very good basis for overall long-term results. Valid results would be based on hours of mining, sometimes GPU crashes with new settings after a while, sometimes many hw errors occur over time, these may not present themselves until many hours of mining.

Cheers!

I test this  1-2 day. This is only SS for send..
With this amd mem tweak setup my 3*vega56 powercolor and 1*vega64LE (all samsung mem) work fine with many algo- I test  CRN-GRAFT- TRTL- MOX-WOW-MSR-LITEv7. any algo have better hr if i change mem tweak setup but with this  setings all algo work fine.

I have a Win 10 - Vega 56 rig

One Hynix card causes a dead GPU when trying 19.4.2, latest TR and have tried both of the suggested Hynix setttings.  Anyone have one that is working for them?   I've tried:

WinAMDTweak.exe --gpu 03 --RAS 24 --RCDRD 19 --RCDWR 4 --RC 35 --RP 13 --RRDS 4 --RRDL 5 --RFC 148 --REF 15600

and

WinAMDTweak.exe --gpu 03 --RAS 22 --RCDRD 17 --RCDWR 4 --RC 35 --RP 13 --RRDS 4 --RRDL 4 --RFC 148 --REF 15600

Both of these cause a dead gpu/restart

try this-

--rcdrd 19 --rcdwr 4 --rc 35 --rp 14 --rrds 4 --rrdl 5 --rfc 148 --REF 15600
or
--rp 14 --rc 42 --rfc 156 --rrd 3 --rcdrd 18 --rcdwr 3

not best but work

iff any have good setup for vega56/stockbios/hynix pls send ..

I got vega 56 strix hynix and i cant go more than 1800hs.. What is ur best results guys? Any good setup for this bad cards?

my 6*veg56 nitro+(hynix) work fine with this setings (+2000hr)- 1 card give 1 error every 1 hour but work fine. itest this with more algo.

WinAMDTweak.exe --rp 14 --rc 35 --rfc 178 --rrds 4 --rrdl 5 --rcdrd 19 --rcdwr 5 --REF 15600

if not work try add --RAS 44(gives stability and lower HR) or chg --rfc 178 to --rfc 188(lower HR)

best card  have gpu= 1440mhz/866mv mem=935mhz/868mv
bad card have   gpu= 1400mhz/870mv mem=895mhz/875mv and give error every 1hour
http://imgbox.com/4Ast1m6p

Will be interesting to know if --REF 15600 still works for you in the summer months.