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

[Anwil]

Gigabyte RX VEGA 56 Hynix memory  (1220/887mv   900/905mv) uptime  2 days.
cr/tartl   19.8 - 20.3
WinAMDTweak.exe --rcdrd 19 --rcdwr 5 --rc 35 --rp 13 --rrds 4 --rrdl 5 --rfc 168 --REF 15600     

[1714306903]

1714306903

[1714306903]

1714306903

[1714306903]

1714306903

[1714306903]

1714306903

[1714306903]

1714306903

[1714306903]

1714306903

[seefatlow]

Hah, it works. Thanks for the tips for getting it to start properly.

Started with todxx's timings and yup, went from ~19.5 to 21.3+. Nice. I'll take that. Let's leave it for a while and see if it can sustain the timings in the long run.

I had the same problem getting it to start. How did you get it to work? I tried running under admin, still not working.

Go to the properties of the exe file and set it to start as admin (iirc). Then make a bat file with your parameters.

It might still only flash on your screen when you run it, but the timings are actually applied.

Got it. Thanks! The screen flashes in an instant. Any ideas how to get a readout from --current?

[Zorg33]

Got it. Thanks! The screen flashes in an instant. Any ideas how to get a readout from --current?

Add pause to the batfile as I already said.
Getting a little kindergartenish.

[seefatlow]

Got it. Thanks! The screen flashes in an instant. Any ideas how to get a readout from --current?

Add pause to the batfile as I already said.
Getting a little kindergartenish.

Dude, if it was that simple, i wouldn't be asking. Running the WinAMDTweak.exe file opens a new window that closes immediately in a flash. It cannot be paused by the pause function in the batch file.

This below in the bat file doesn't work.

c:\<FULL PATH TO THE TOOL>\WinAMDTweak.exe --current
pause

[Bojcha]

Got it. Thanks! The screen flashes in an instant. Any ideas how to get a readout from --current?

Add pause to the batfile as I already said.
Getting a little kindergartenish.

Dude, if it was that simple, i wouldn't be asking. Running the WinAMDTweak.exe file opens a new window that closes immediately in a flash. It cannot be paused by the pause function in the batch file.

This below in the bat file doesn't work.

c:\<FULL PATH TO THE TOOL>\WinAMDTweak.exe --current
pause

Admin Rights. Only way is to open cmd as admin and type it manually.

[carlosmonaco]

Thanks alot Eliovp for your efforts for the windows version. I have been learning this vague ubuntu thing for the last week until you released windows version.
Now, for the love of god, this 18.50 driver with Ubuntu 18.04 is installable? As a desktop ubuntu NOT headless (i use the machine mainly to experiment on mining, and hell yes this linux is WAY stable than windows.) Can anyone tell me how i can get 18.50 GPU pro running? i tried installing amdgpu install and opencl only from amdgpu pro, tried alot and mostly end up in a stuck boot.


2nd issue! Windows, mate, i tried changing timings simply tRRD of elpida while its running..... (mining Graft with TRD 0.4.3) it succeds in changing tRRd to 4 for example, but when i use --current its back to old value (5) also tCKE/Pulse of Hynix AJR is wierd! 13/13 underload.... oO same with other wierd readings like bus_turn. are these things getting changed dynamically? also my gpus give memory errors while mining so that might explain failure to alter timings?

I recommend to go online and search for a simple tutorial on how to install amdgpu drivers. It's not that hard but i can imagine if it's your first time, that it might be a bit tricky.

I'm working on something to make this process easier.  ;-)

Great tool. I have been playing a bit mainly with list timings in previous post and have found something that works for each type of card I have, V56 V56 flashed to V64 and V64.

Thank you soo much for releasing this tool.
Do you also have an XMR address, would be an idea to post up and maybe we can all throw some hashpower your way aswell.
you should be greatly rewarded for your work.

of topic:
I am also using a hynix memory strap you once posted on my rx550's and it is amazing

Hey Mashy, sorry sir, no XMR address.
You can always use a shift service if you want to tip.
coinswap, changelly.. etc..

Using TRM 0.4.4 and last driver ;

Working great with my vega 64/56 reference card ! Around 2200h/s avg..


But summer is coming , is there a way to decrease consumption with this this tool ?

For now my card are running at 1407/1100 or 1407/950 at 875mV for most of them and 900-930 mV for bad ones.

Who knows.. perhaps something is coming  

Something in mind ? ^^

[clousian]

Still no valid values for us un-bios-modded folks on Polaris cards?

Yeah but not everyone wants to bios mod they cards. Kinda interested in modding my cards on the fly with this tool.
Most people are not using this app for Polaris because good timing straps are already available for Samsung, Elpida and Hynix memory. I think Micron is the only strap not available to my knowledge. If you are using Polaris cards with these memory, just flash your bios using common known good straps.

[kaiokenx10]

Hah, it works. Thanks for the tips for getting it to start properly.

Started with todxx's timings and yup, went from ~19.5 to 21.3+. Nice. I'll take that. Let's leave it for a while and see if it can sustain the timings in the long run.

I had the same problem getting it to start. How did you get it to work? I tried running under admin, still not working.

Go to the properties of the exe file and set it to start as admin (iirc). Then make a bat file with your parameters.

It might still only flash on your screen when you run it, but the timings are actually applied.

Got it. Thanks! The screen flashes in an instant. Any ideas how to get a readout from --current?

all you have to do is open cmd as admin and navigate to the folder where your memtweak is located

that would be:

cd path to folder

once you got into the folder, you just type in WinAMDTweak.exe --current

[BillGatesIII]

Hey Mashy, sorry sir, no XMR address.
You can always use a shift service if you want to tip.
coinswap, changelly.. etc..

Get yourself one, I'll be happy to mine some XMR for you

[Eliovp]

Hey Mashy, sorry sir, no XMR address.
You can always use a shift service if you want to tip.
coinswap, changelly.. etc..

Get yourself one, I'll be happy to mine some XMR for you

Haha, ok ok, you convinced me ;-)

Check OP

And thank you!

[Kerppa77]

I have now run ~72 hours w/o any issues with 24 Vega64 GPUs(good and bad ones) mixed in 4 separate rigs ... CNR with core 1407Mhz /860-970mV , Mem 1080-1100Mhz/860-970mV.

I would say this is pretty stable setup. Pool side 24h avg hashrate fluctuates in the range 54,1-54,4 kH/s

--CL 19 --RAS 28 --RCDRD 12 --RCDWR 5 --RC 44 --RP 12 --RRDS 3 --RRDL 3 --REF 15600 --RFC 250

I am happy and have in my mind... "dont touch when things go smoothly".

[pbfarmer]

I have now run ~72 hours w/o any issues with 24 Vega64 GPUs(good and bad ones) mixed in 4 separate rigs ... CNR with core 1407Mhz /860-970mV , Mem 1080-1100Mhz/860-970mV.

I would say this is pretty stable setup. Pool side 24h avg hashrate fluctuates in the range 54,1-54,4 kH/s

--CL 19 --RAS 28 --RCDRD 12 --RCDWR 5 --RC 44 --RP 12 --RRDS 3 --RRDL 3 --REF 15600 --RFC 250

I am happy and have in my mind... "dont touch when things go smoothly".

That’s an insane voltage range, and I can’t fathom how 970mv could possibly be necessary. Between 9 64s and a flashed 56, I run avg 834mv, none over 850, for similar clocks/timings.  That puts me @ 178w per GPU at the wall.

At 970, you’re running at least 35% higher power, meaning over 240w per card.

It sounds like you have some ppt issues or something, maybe you’re running your SOC @ 1200mhz unnecessarily (it should be 1107 @ your clocks.)  Although, even then, you prob shouldn’t need much more than 925mv.

[mingdao]

Anyone cares to share some tunes for Vega FEs?
My test one runs as follows, does stable 2.2k on CNR:

GPU 0:  Vega 10 XTX [Radeon Vega Frontier Edition]      Memory state: 0x11C (1000MHz)
Timing 1          CL: 17          RAS: 33         RCDRD: 16       RCDWR: 10
Timing 2          RCAb (RC): 47   RCPb (RC): 47   RPAb (RP): 14   RPPb (RP): 14
Timing 3          RRDS: 4         RRDL: 6         RTP: 5
Timing 4          FAW: 16
Timing 5          CWL: 6          WTRS: 4         WTRL: 9
Timing 6          WR: 16
Timing 7          RREFD: 8
Timing 8          RDRDDD: 3       RDRDSD: 2       RDRDSC: 1       RDRDSCL: 3
Timing 9          WRWRDD: 3       WRWRSD: 2       WRWRSC: 1       WRWRSCL: 3
Timing 10         WRRD: 1         RDWR: 17
Timing 12         REF: 3900
Timing 13         MRD: 8          MOD: 15
Timing 14         XS: 270
Timing 16         XSMRS: 8
Timing 17         PD: 8   CKSRE: 10       CKSRX: 10
Timing 20         RFCPB: 160      STAG: 7
Timing 21         XP: 8   CPDED: 2        CKE: 8
Timing 22         RDDATA: 16      WRLAT: 4        RDLAT: 14       WRDATA: 1
Timing 23         CKESTAG: 2
RFC Timing        RFC: 350

[carlosmonaco]

I have now run ~72 hours w/o any issues with 24 Vega64 GPUs(good and bad ones) mixed in 4 separate rigs ... CNR with core 1407Mhz /860-970mV , Mem 1080-1100Mhz/860-970mV.

I would say this is pretty stable setup. Pool side 24h avg hashrate fluctuates in the range 54,1-54,4 kH/s

--CL 19 --RAS 28 --RCDRD 12 --RCDWR 5 --RC 44 --RP 12 --RRDS 3 --RRDL 3 --REF 15600 --RFC 250

I am happy and have in my mind... "dont touch when things go smoothly".

That’s an insane voltage range, and I can’t fathom how 970mv could possibly be necessary. Between 9 64s and a flashed 56, I run avg 834mv, none over 850, for similar clocks/timings.  That puts me @ 178w per GPU at the wall.

At 970, you’re running at least 35% higher power, meaning over 240w per card.

It sounds like you have some ppt issues or something, maybe you’re running your SOC @ 1200mhz unnecessarily (it should be 1107 @ your clocks.)  Although, even then, you prob shouldn’t need much more than 925mv.

Oh ! I'm far above that number too.
Could you share your PPT tables  and settings ?
Thanks

[GordoLui]

Hah, it works. Thanks for the tips for getting it to start properly.

Started with todxx's timings and yup, went from ~19.5 to 21.3+. Nice. I'll take that. Let's leave it for a while and see if it can sustain the timings in the long run.

I had the same problem getting it to start. How did you get it to work? I tried running under admin, still not working.

Go to the properties of the exe file and set it to start as admin (iirc). Then make a bat file with your parameters.

It might still only flash on your screen when you run it, but the timings are actually applied.

Got it. Thanks! The screen flashes in an instant. Any ideas how to get a readout from --current?

all you have to do is open cmd as admin and navigate to the folder where your memtweak is located

that would be:

cd path to folder

once you got into the folder, you just type in WinAMDTweak.exe --current

Not sure why my pause .bat files did not work, but just tried doing this and im able to use WinAMDTweak and change the setting from there. Now i can get to work on this.

[pbfarmer]

I have now run ~72 hours w/o any issues with 24 Vega64 GPUs(good and bad ones) mixed in 4 separate rigs ... CNR with core 1407Mhz /860-970mV , Mem 1080-1100Mhz/860-970mV.

I would say this is pretty stable setup. Pool side 24h avg hashrate fluctuates in the range 54,1-54,4 kH/s

--CL 19 --RAS 28 --RCDRD 12 --RCDWR 5 --RC 44 --RP 12 --RRDS 3 --RRDL 3 --REF 15600 --RFC 250

I am happy and have in my mind... "dont touch when things go smoothly".

That’s an insane voltage range, and I can’t fathom how 970mv could possibly be necessary. Between 9 64s and a flashed 56, I run avg 834mv, none over 850, for similar clocks/timings.  That puts me @ 178w per GPU at the wall.

At 970, you’re running at least 35% higher power, meaning over 240w per card.

It sounds like you have some ppt issues or something, maybe you’re running your SOC @ 1200mhz unnecessarily (it should be 1107 @ your clocks.)  Although, even then, you prob shouldn’t need much more than 925mv.

Oh ! I'm far above that number too.
Could you share your PPT tables  and settings ?
Thanks

I'm on linux, and only use PPT (no additional settings.)  But you won't be able use my PPTs on windows.

[rebelmalangi]

I know this post is for Vega, but can anyone timing for Rx 580 (appreciate) or process how-to play with the tool to get the best time for the cards. Github is with commands which we are not familiar.

[Zorg33]

For those having issues with running the tool as admin and from a bat file. Please keep in mind that you have to run the bat file itself as admin, otherwise the tool will open in another cmd instance and you won't see any otput (if run as admin is set in the tool's properties). Also you must use full path to the tool or first cd to the folder because admin cmd always starts from system32 folder.
I hope there won't be any more question in this matter as it makes this thread very noisy.

[Kerppa77]

I have now run ~72 hours w/o any issues with 24 Vega64 GPUs(good and bad ones) mixed in 4 separate rigs ... CNR with core 1407Mhz /860-970mV , Mem 1080-1100Mhz/860-970mV.

I would say this is pretty stable setup. Pool side 24h avg hashrate fluctuates in the range 54,1-54,4 kH/s

--CL 19 --RAS 28 --RCDRD 12 --RCDWR 5 --RC 44 --RP 12 --RRDS 3 --RRDL 3 --REF 15600 --RFC 250

I am happy and have in my mind... "dont touch when things go smoothly".

That’s an insane voltage range, and I can’t fathom how 970mv could possibly be necessary. Between 9 64s and a flashed 56, I run avg 834mv, none over 850, for similar clocks/timings.  That puts me @ 178w per GPU at the wall.

At 970, you’re running at least 35% higher power, meaning over 240w per card.

It sounds like you have some ppt issues or something, maybe you’re running your SOC @ 1200mhz unnecessarily (it should be 1107 @ your clocks.)  Although, even then, you prob shouldn’t need much more than 925mv.

Oops... I made a quick check with HWtool and yes... I really have SOC max @ 1199Mhz. I have used the Hellae 920mV PPT and then adjusted with Overdrive Ntool the Core P7 and Mem P3. I havent touch the SOC. After memory timings my rigs are pulling like 1420-1500w from the wall per rig. My rigs run like 150w w/o GPUs and after OverdriveNtool the wattage is ~250w before the miner starts and all the GPU led lights are fully on.
This sounds to me like 1270 to 1350w with 6 GPU and on avg. like 210 - 225w/GPU. :O

Thanks for pointing out the SOC. I need to drop to 1107.

[kerney666]

I have now run ~72 hours w/o any issues with 24 Vega64 GPUs(good and bad ones) mixed in 4 separate rigs ... CNR with core 1407Mhz /860-970mV , Mem 1080-1100Mhz/860-970mV.

I would say this is pretty stable setup. Pool side 24h avg hashrate fluctuates in the range 54,1-54,4 kH/s

--CL 19 --RAS 28 --RCDRD 12 --RCDWR 5 --RC 44 --RP 12 --RRDS 3 --RRDL 3 --REF 15600 --RFC 250

I am happy and have in my mind... "dont touch when things go smoothly".

That’s an insane voltage range, and I can’t fathom how 970mv could possibly be necessary. Between 9 64s and a flashed 56, I run avg 834mv, none over 850, for similar clocks/timings.  That puts me @ 178w per GPU at the wall.

At 970, you’re running at least 35% higher power, meaning over 240w per card.

It sounds like you have some ppt issues or something, maybe you’re running your SOC @ 1200mhz unnecessarily (it should be 1107 @ your clocks.)  Although, even then, you prob shouldn’t need much more than 925mv.

Oops... I made a quick check with HWtool and yes... I really have SOC max @ 1199Mhz. I have used the Hellae 920mV PPT and then adjusted with Overdrive Ntool the Core P7 and Mem P3. I havent touch the SOC. After memory timings my rigs are pulling like 1420-1500w from the wall per rig. My rigs run like 150w w/o GPUs and after OverdriveNtool the wattage is ~250w before the miner starts and all the GPU led lights are fully on.
This sounds to me like 1270 to 1350w with 6 GPU and on avg. like 210 - 225w/GPU. :O

Thanks for pointing out the SOC. I need to drop to 1107.

When I tried to scribble down a new Vega tuning guide to include in the TRM 0.4.4 release, I was debating if I should have mentioned this or not. I opted not to, that text was long enough already anyway.

However, I know more than a few skilled and knowledgeable Vega tuners that ALL have been stuck with old PPTs from the fall of 2017 (brnsted, Hellae) that for some reason all have 1200 for SoC. At the time, I guess it was considered highly beneficial for stability. Fast forward a few years to times where you don't make $10/day/Vega and power draw is a crucial part of your setup, this trade-off no longer holds.

In the common Google sheet (https://docs.google.com/spreadsheets/d/1-rhYsaRXO1ahk3PyrEgT9gXzs7ImAzh-sbqtgwy8HQg) used for PPT tweaking you can edit the SOC clk value but guess what: it doesn't change any nrs in the PPT, the cells aren't connected(!!!). You have to change the SOC clk value in cell AE26, then manually copy cells AG-AI:26 to T-V:11. You can probably guess why I may sound a little pissed off  .