AMD Mem Tweak XL - Read/modify timings/pp/straps on the fly - page 26.

pbfarmer

member

Activity: 340

Merit: 29

Quote from: kamisama233 on May 28, 2019, 08:59:04 AM

Quote from: pbfarmer on May 24, 2019, 09:47:34 PM

For anyone else using the old PBE one-click samsung straps (aka generaleramon's ubermix 3.2) on RX 4/5XX 8gb GPUs, and seeing subpar ETH h/rs due to the tight timings limiting mem clock frequencies... I've been able to get my samsung 580-8s more in-line w/ the microns (~32.5 MH/s) by loosening a couple settings which then allow raising the mem clock (combined w/ a REF increase i use on all my 580s):

--REF 10 --RC 70 --RFC 157 (from REF 5, RC 65, RFC 136)
mclock 2145-2185 (from 2085-2145)

other relevant bits:

cclock 1175, 806mv, phoenixminer, and ubuntu 18.04 + amdgpu-pro 18.50 (though windows should behave similarly)

You could prob get to ~33+ w/ a 1200-1250+ cclock, while I may have slightly overshot on the loosened timings. But as usual, I'm more interested in efficiency than max h/rs.
Also, mclock is still a bit limited, as i believe CL (currently 21) would need to be relaxed to get to 2200+ MHz -- which as i understand, can't be done w/ the tool atm.

excellent, how can you lower your 580's voltage to such a low level with a 1175 frequency, core and memory voltage all at 806mv, that's impressive, how to lower, thanks

No tricks - i just set them that way (I use ppts, but that's prob not necessary.) Not all cards can do the same - my range on 580s for ethash is 800-837mv, though I haven't been super-detailed on tuning all of them, so the range could be lower/tighter.

Also, proper cooling can help a lot... More heat = more power use, and not just because of higher fan speed. I have mostly sapphire nitros, w/ a fair amount of spacing between GPUs (enough for ~120mm fan between each) and some decent airflow which keeps my temps in the low-mid 40s mostly.

tallman_ay

newbie

Activity: 5

Merit: 0

https://prnt.sc/nuo8i5
I dont know why every time I try to run the GUi I get this error msg any help please

Mashy81

jr. member

Activity: 225

Merit: 1

Quote from: MaxHa$h on May 28, 2019, 09:45:56 AM

Quote from: Mashy81 on May 28, 2019, 02:42:29 AM

Now getting 555hs with each of my rx550's.
1180 core 1930 memory

2gb or 4gb?

What timings did you use?

2gb.
They are bios modded, memory strap from somewhere on the Net (530hs)
Just been playing around with different numbers that I can now access with this tool. REF 50 rfc 120 rc 56. Are a few
Will upload the full timing when I'm at my pc

MaxHa$h

newbie

Activity: 37

Merit: 0

Quote from: Mashy81 on May 28, 2019, 02:42:29 AM

Now getting 555hs with each of my rx550's.
1180 core 1930 memory
😁

2gb or 4gb?

What timings did you use?

kamisama233

newbie

Activity: 42

Merit: 0

Quote from: pbfarmer on May 24, 2019, 09:47:34 PM

For anyone else using the old PBE one-click samsung straps (aka generaleramon's ubermix 3.2) on RX 4/5XX 8gb GPUs, and seeing subpar ETH h/rs due to the tight timings limiting mem clock frequencies... I've been able to get my samsung 580-8s more in-line w/ the microns (~32.5 MH/s) by loosening a couple settings which then allow raising the mem clock (combined w/ a REF increase i use on all my 580s):

--REF 10 --RC 70 --RFC 157 (from REF 5, RC 65, RFC 136)
mclock 2145-2185 (from 2085-2145)

other relevant bits:

cclock 1175, 806mv, phoenixminer, and ubuntu 18.04 + amdgpu-pro 18.50 (though windows should behave similarly)

You could prob get to ~33+ w/ a 1200-1250+ cclock, while I may have slightly overshot on the loosened timings. But as usual, I'm more interested in efficiency than max h/rs.
Also, mclock is still a bit limited, as i believe CL (currently 21) would need to be relaxed to get to 2200+ MHz -- which as i understand, can't be done w/ the tool atm.

excellent, how can you lower your 580's voltage to such a low level with a 1175 frequency, core and memory voltage all at 806mv, that's impressive, how to lower, thanks

Xform

newbie

Activity: 5

Merit: 0

Hello!

CLI version of amdmemtweak do nothing on my Polaris (RX584). GUI vesrsion is all OK. For example:

Code:

C:\Mining\MemTweak>winamdtweak --REF 30

C:\Mining\MemTweak>

No messages, no actions.

Windows 10 LTSB, driver is "Blockchain"

What is wrong?

Mashy81

jr. member

Activity: 225

Merit: 1

Now getting 555hs with each of my rx550's.
1180 core 1930 memory
😁

kamisama233

newbie

Activity: 42

Merit: 0

Quote from: Eliovp on May 22, 2019, 08:24:11 PM

Quote from: pbfarmer on May 22, 2019, 04:14:24 PM

@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 Grin

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!

can you add support for lexa card in cli version

pbfarmer

member

Activity: 340

Merit: 29

Quote from: Collycoin on May 27, 2019, 05:09:06 AM

For those running Ubuntu 18.04 & AMD Cards.

How are you modifying the core & mem clocks and voltages of AMD cards?
In Windows AMD Mem Tweak allows these to be changed in the GUI but I can't find a way to do it in Ubuntu in AMD Mem Tweak or any other way.
I've got thirsty Vega 56s that desperately need to be downvolted (otherwise my wife will kill me on the electricity bills!).

Please help!

I just use PPTs, but you can also use the sysfs interface:

https://wiki.archlinux.org/index.php/AMDGPU#Overclocking

Miner Addict

newbie

Activity: 31

Merit: 0

Quote from: Miner Addict on May 27, 2019, 12:12:45 PM

Still, cannot apply setting at GUI version with start up

GUI open with default setting ?

Anyone got a tutorial ?

Because at Polaris, REFF is Important

Miner Addict

newbie

Activity: 31

Merit: 0

Still, cannot apply setting at GUI version with start up

GUI open with default setting ?

Anyone got a tutorial ?

Collycoin

newbie

Activity: 17

Merit: 0

For those running Ubuntu 18.04 & AMD Cards.

How are you modifying the core & mem clocks and voltages of AMD cards?
In Windows AMD Mem Tweak allows these to be changed in the GUI but I can't find a way to do it in Ubuntu in AMD Mem Tweak or any other way.
I've got thirsty Vega 56s that desperately need to be downvolted (otherwise my wife will kill me on the electricity bills!).

Please help!

rednoW

legendary

Activity: 1510

Merit: 1003

Quote from: Alantre on May 26, 2019, 08:46:25 PM

What are peoples getting for radeon vii for eth?
I saw 3100h+ CNr @ 170w(I am assuming from wall?)

3100 will consume 180-220watt depending on silicon lottery. cn-r takes some advantage from memory overclock and timings on radeon 7 but this impact is small.
eth is 80-82mhs at 200-220watt and no boost from memory clocks above 940mhz and memory timings.
Only core clock matters

Alantre

newbie

Activity: 18

Merit: 0

What are peoples getting for radeon vii for eth?
I saw 3100h+ CNr @ 170w(I am assuming from wall?)

WhackOBill

jr. member

Activity: 73

Merit: 3

Quote from: pbfarmer on May 26, 2019, 12:54:49 AM

That's interesting - I had always thought I was limited to ~2K on ubermix as well, but during this exercise of finding a way to get to higher clocks, I found out I could get to ~2.1K w/ no timings changes. I guess the 2K limit (at least for me) was more the case under CN - ethash must be a bit easier on the mem.

I'm also surprised you can do REF 30 - I can do higher refs w/ CN, but for ethash, I'm seeing a couple percent errors already at REF 10 (def depends on ambient tho.)

Maybe we're using different ubermix versions? iirc, a couple different ones were included in PBE over time. I also recently noticed generaleramon had published later versions (even past 3.2,) one of which was suggested may be ideal for 2150.

I think I am running Ubermix 3.1 but it's been a while. Your original values for RC and RFC match mine. At REF 30, I have 1 bad hash in about 24 hours. On my Hynix cards, I can't run beyond REF 18 without too many bad hashes.

This reminds me of the days when a group of us using ATT6300 PC's (8086 machines) would lower the dram refresh rate from that set in BIOS to gain about 12% in processing speed. Of course, our CPUs were running at 8MHz so that was a big deal for us.

pbfarmer

member

Activity: 340

Merit: 29

Quote from: WhackOBill on May 25, 2019, 06:15:11 PM

Quote from: pbfarmer on May 24, 2019, 09:47:34 PM

For anyone else using the old PBE one-click samsung straps (aka generaleramon's ubermix 3.2) on RX 4/5XX 8gb GPUs, and seeing subpar ETH h/rs due to the tight timings limiting mem clock frequencies... I've been able to get my samsung 580-8s more in-line w/ the microns (~32.5 MH/s) by loosening a couple settings which then allow raising the mem clock (combined w/ a REF increase i use on all my 580s):

--REF 10 --RC 70 --RFC 157 (from REF 5, RC 65, RFC 136)
mclock 2145-2185 (from 2085-2145)

other relevant bits:

cclock 1175, 806mv, phoenixminer, and ubuntu 18.04 + amdgpu-pro 18.50 (though windows should behave similarly)

You could prob get to ~33+ w/ a 1200-1250+ cclock, while I may have slightly overshot on the loosened timings. But as usual, I'm more interested in efficiency than max h/rs.
Also, mclock is still a bit limited, as i believe CL (currently 21) would need to be relaxed to get to 2200+ MHz -- which as i understand, can't be done w/ the tool atm.

My Samsung memory is weak, only able to run at 1960 with Ubermix. I was able to bump to 2000 with your suggestion along with a REF 30 and gained 0.7MHs. I would definitely have to loosen Column Address Strobe Latency to go higher.

That's interesting - I had always thought I was limited to ~2K on ubermix as well, but during this exercise of finding a way to get to higher clocks, I found out I could get to ~2.1K w/ no timings changes. I guess the 2K limit (at least for me) was more the case under CN - ethash must be a bit easier on the mem.

I'm also surprised you can do REF 30 - I can do higher refs w/ CN, but for ethash, I'm seeing a couple percent errors already at REF 10 (def depends on ambient tho.)

Maybe we're using different ubermix versions? iirc, a couple different ones were included in PBE over time. I also recently noticed generaleramon had published later versions (even past 3.2,) one of which was suggested may be ideal for 2150.

WhackOBill

jr. member

Activity: 73

Merit: 3

Quote from: pbfarmer on May 24, 2019, 09:47:34 PM

For anyone else using the old PBE one-click samsung straps (aka generaleramon's ubermix 3.2) on RX 4/5XX 8gb GPUs, and seeing subpar ETH h/rs due to the tight timings limiting mem clock frequencies... I've been able to get my samsung 580-8s more in-line w/ the microns (~32.5 MH/s) by loosening a couple settings which then allow raising the mem clock (combined w/ a REF increase i use on all my 580s):

--REF 10 --RC 70 --RFC 157 (from REF 5, RC 65, RFC 136)
mclock 2145-2185 (from 2085-2145)

other relevant bits:

cclock 1175, 806mv, phoenixminer, and ubuntu 18.04 + amdgpu-pro 18.50 (though windows should behave similarly)

You could prob get to ~33+ w/ a 1200-1250+ cclock, while I may have slightly overshot on the loosened timings. But as usual, I'm more interested in efficiency than max h/rs.
Also, mclock is still a bit limited, as i believe CL (currently 21) would need to be relaxed to get to 2200+ MHz -- which as i understand, can't be done w/ the tool atm.

My Samsung memory is weak, only able to run at 1960 with Ubermix. I was able to bump to 2000 with your suggestion along with a REF 30 and gained 0.7MHs. I would definitely have to loosen Column Address Strobe Latency to go higher.

lstrike

newbie

Activity: 31

Merit: 0

Yes... there are issues with saving/opening profiles, also applying changes while mining...

fury420

newbie

Activity: 51

Merit: 0

I seem to have found a conflict with HWInfo, might be worth mentioning to save others hassle troubleshooting.

It seems if HWinfo has been opened in sensors mode it prevents AMDMemTweak from applying any timing changes until the system is rebooted.
Even after quitting HWinfo any changes in the timing sections still will not apply, they just revert back to the previous value when I press Apply.
Changes to the Clock/Voltage/Fan Control sections still work, just not the timings or REF adjustments.

Any chance of adding a GPU Memory Errors counter to the tool somewhere so I can ditch HWinfo?
Or do you know of a tool that will show memory errors and does not conflict?

rednoW

legendary

Activity: 1510

Merit: 1003

2 Eliovp

Hi! Seems I found one more bug in beta 5 and beta 6.
Memory frequency for Radeon VII is not saved/loaded correctly from profiles.
In beta 4 all was ok.
Also would like a feature to have an option to apply only freq/voltage/fans without memory timings loaded from profile.
Thanks for your great work!

Topic: AMD Mem Tweak XL - Read/modify timings/pp/straps on the fly - page 26. (Read 58833 times)