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

[Marvell2]

I’m trying to get that 50 mh on vega 56. I got up to 46 right now. Do I need to bios flash it to a 64?? If so is it still possible if I got Hynix memory?  All I see is Samsung memory bios flash.

Gave up on my 4 56 vegas and my frontier editions, fe should be able to get close to 60 MHz eth with its 16gb hbm, it gets 45 MHz with just mem boost to 1100.

My 56 ref cards work great at 50 MHz but they were flashed to 64

Just Gona eBay the 56 non ref and fe’s

[1679801606]

1679801606

[1679801606]

1679801606

[1679801606]

1679801606

[Holiidayz]

what kind of drivers would work well for this to improve it for flashed hynix 56, hynix 470/570/580, and samsung 580?
edit: on windows 10 build

[goxed]

I’m trying to get that 50 mh on vega 56. I got up to 46 right now. Do I need to bios flash it to a 64?? If so is it still possible if I got Hynix memory?  All I see is Samsung memory bios flash.

Gave up on my 4 56 vegas and my frontier editions, fe should be able to get close to 60 MHz eth with its 16gb hbm, it gets 45 MHz with just mem boost to 1100.

My 56 ref cards work great at 50 MHz but they were flashed to 64

Just Gona eBay the 56 non ref and fe’s

Why do you think FE will get 60mh on ethash? MHz for MHz, FE HBM2 is generally slower than RX vega HBM2.

[Zerberfert]

Just a quick question. Is there a brand name for the Vega's to stay away from. It seems I see more problems with Power Color than any other brand. At least from my research. I want to play with the Amd Mem Tweek tool soon, so about to bulk up on Vega's. Thanks in advance.

Zerb

[QuirkSilver]

Eliovp, so only with 19.4.x driver i could get the tool to to run, with 560/570/580 Tool is prompt when i tested, got hash increase.
However, after a disconnect and miner restating or anything similar, i can't apply any timings not sure why, also, this 19.x drivers doesn't play well with most miners.
I also lost temperature control with overdriven or your tool, on those drivers.
By newish drivers as a requirement, which do you can do? 18.6.1/18.7.1 ?
also, thats a technical question, why do memory in action, change Dram and Dram timing 2 also EDC
I'm on windows 10 pro 1803

[Eliovp]

Eliovp, so only with 19.4.x driver i could get the tool to to run, with 560/570/580 Tool is prompt when i tested, got hash increase.
However, after a disconnect and miner restating or anything similar, i can't apply any timings not sure why, also, this 19.x drivers doesn't play well with most miners.
I also lost temperature control with overdriven or your tool, on those drivers.
By newish drivers as a requirement, which do you can do? 18.6.1/18.7.1 ?
also, thats a technical question, why do memory in action, change Dram and Dram timing 2 also EDC
I'm on windows 10 pro 1803

There's like 50 different cases to check, driver version, overdrive compatibility, gpu's, ...  if i had to add support for every driver, gpu, overdrive api, i'd go insane.

Regarding dram, dram values are calculated upon boot based on clocks and vbios values.

So for everyone else with "similar" issues.

In case you experience issues with the GUI version, update drivers, use DDU to remove old ones first.
If that still gives you issues, simply use the CLI version which does not have overdrive support (Clock/Fan/.. control)

That said.

I will release a new version (probably today) with Fiji support, some new timings, some bug fixes, memory manufacturer for HBM2, etc...

Cheers!

[Eliovp]

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

[doktor83]

RxBoost™

[dragonmike]

Added Refresh Rate timing (RX boost  Wink)

RXboost... Is that Claymore's thing?
What do you actually get out of that? Does it run independently to the timing straps? Don't want to f@ck anything up that's seemingly stable!

[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.

[ganzocrypt]

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

I tried to set the REF and RTC as you mention on my RX580 8Gb Hynex but as soon as I push apply, it gets back to original values, any idea?
thx

[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

[pigfrown]

Built the latest linux version from github but get the following error message

Code:

Cannot find DRI instance for pci:0000:01:00.0

Same message regardless of what I pass with --i , e.g.

Code:

root@minivega:/home/miner/scratch/amdmemorytweak/linux# ./amdmemtweak --i 0 --current
Cannot find DRI instance for pci:0000:01:00.0
root@minivega:/home/miner/scratch/amdmemorytweak/linux# ./amdmemtweak --i 1 --current
Cannot find DRI instance for pci:0000:01:00.0
root@minivega:/home/miner/scratch/amdmemorytweak/linux# ./amdmemtweak --i 2 --current
Cannot find DRI instance for pci:0000:01:00.0

System details:

Code:

root@minivega:/dev/dri# uname -a
Linux minivega 4.4.0-131-generic #157-Ubuntu SMP Thu Jul 12 15:51:36 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux
root@minivega:/dev/dri# apt list --installed | grep amdgpu-pro

WARNING: apt does not have a stable CLI interface. Use with caution in scripts.

amdgpu-pro-core/unknown,now 18.10-572953 all [installed,automatic]
clinfo-amdgpu-pro/unknown,now 18.10-572953 amd64 [installed]
libopencl1-amdgpu-pro/unknown,now 18.10-572953 amd64 [installed,automatic]
opencl-amdgpu-pro/unknown,now 18.10-572953 amd64 [installed]
opencl-amdgpu-pro-icd/unknown,now 18.10-572953 amd64 [installed,automatic]
opencl-orca-amdgpu-pro-icd/unknown,now 18.10-572953 amd64 [installed]

Anyone else got the same problem, or do I need to update amdgpu-pro?

[Eliovp]

Built the latest linux version from github but get the following error message

Code:

Cannot find DRI instance for pci:0000:01:00.0

Same message regardless of what I pass with --i , e.g.

Code:

root@minivega:/home/miner/scratch/amdmemorytweak/linux# ./amdmemtweak --i 0 --current
Cannot find DRI instance for pci:0000:01:00.0
root@minivega:/home/miner/scratch/amdmemorytweak/linux# ./amdmemtweak --i 1 --current
Cannot find DRI instance for pci:0000:01:00.0
root@minivega:/home/miner/scratch/amdmemorytweak/linux# ./amdmemtweak --i 2 --current
Cannot find DRI instance for pci:0000:01:00.0

System details:

Code:

root@minivega:/dev/dri# uname -a
Linux minivega 4.4.0-131-generic #157-Ubuntu SMP Thu Jul 12 15:51:36 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux
root@minivega:/dev/dri# apt list --installed | grep amdgpu-pro

WARNING: apt does not have a stable CLI interface. Use with caution in scripts.

amdgpu-pro-core/unknown,now 18.10-572953 all [installed,automatic]
clinfo-amdgpu-pro/unknown,now 18.10-572953 amd64 [installed]
libopencl1-amdgpu-pro/unknown,now 18.10-572953 amd64 [installed,automatic]
opencl-amdgpu-pro/unknown,now 18.10-572953 amd64 [installed]
opencl-amdgpu-pro-icd/unknown,now 18.10-572953 amd64 [installed,automatic]
opencl-orca-amdgpu-pro-icd/unknown,now 18.10-572953 amd64 [installed]

Anyone else got the same problem, or do I need to update amdgpu-pro?

I just pushed a new CLI version both for windows and Linux, can you please try that one. (there's a pre-built one in releases)

Cheers

[dragonmike]

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

Hey Livada, nice results.
Are these on reference 64's?
What clocks were you running them at per algo?
2.48 on CN/r is nice...

[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

Hey Livada, nice results.
Are these on reference 64's?
What clocks were you running them at per algo?
2.48 on CN/r is nice...

GPU4 is Vega 64LE - GPU 1-2 REferent Vega 56@64 bios (+2300hr on CNR)
All algo run with identical Mem tweak.
Vega 64LE clock :
http://imgbox.com/2xmtiUyw
Vega 56 clock :
http://imgbox.com/TzsvzlCa

[jimmyD30]

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!

[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.

[joseph32]

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

Hey Livada, nice results.
Are these on reference 64's?
What clocks were you running them at per algo?
2.48 on CN/r is nice...

Sorry, but 1 accepted and a 3 minutes run shows nothing. These results with 24 hours would be nice, but I dont bet on it.

[SamAlackass]

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).