Need absolute core clock for nvidia

[thaneb]

Hello there, After watching this great info about the new absolute core clock tweak for nvidia gpus available in hiveos, I really would like to get this option on gminer windows (got a 3060, so i cannot simply jump into hiveos) It seems to drastically reduce power consumption, I don't know if others miners have this overclock setting available, but for sure, if the consumption result is the same on windows, it would be an amazing value-added to your software! https://www.youtube.com/watch?v=7sqtUEWDTTs

thread related at hiveos: https://forum.hiveos.farm/t/locked-core-clock-speed-is-much-better-than-power-limit-why-is-not-included-by-default/29803

[sebek72]

Yes, this would be a great feature (like on HiveOS).

[thaneb]

for those interested, I found how to manually tweak in the start.bat with nvidia-smi -i 0 --lock-gpu-clocks=minimuclock,maxclock before the line starting the miner of course. Here's my actual settings, I would appreciate if testing people could share theirs so we can help to improve each others:

reset all core settings

nvidia-smi -rgc

RTX 3070 twin edge zotac - 62.37MH/s mem +1100 (PL 54)

nvidia-smi -i 0 --lock-gpu-clocks=885,885 nvidia-smi -i 1 --lock-gpu-clocks=885,885

RTX 3060 ti dual Asus - 62.43MH/s mem +1100 (PL 61)

nvidia-smi -i 2 --lock-gpu-clocks=1425,1425

RTX 2070 Gaming Z MSI - 43.97MH/s +760 (PL59)

nvidia-smi -i 3 --lock-gpu-clocks=1005,1005

GTX 1660 super Zotac - 31.51MH/s +760 (PL 70)

nvidia-smi -i 4 --lock-gpu-clocks=1050,1050

My last card is a RTX 3060, I got a huge hashrate drop when I try to lock the core clock to the almost always used automatic value (1125), so for now, I stay with only a -500 / +1200 , PL 70 to get 49.76MH/S (Zotac twin edge)

[sebek72]

Do you have to disable afterburner or can you just override its values via this commands?

[thaneb]

I guess you can just remove afterburner from startup app/services, i removed it from my rig very quickly to use nvidia inspector at first, and since gpu and memory clock offsets are usable through gminer, i don't use inspector anymore. The first command reset gpu clock, so you should not have to worry about afterburner. By writing this, it makes me think that I forgot to check how these settings are applied regarding to the p-state, as requested clock's are well applied to all my gpus, I guess it's applied to p0 state, but that's maybe about something like this that I got issue with the rtx 3060..

[wacher74]

Can I set memory offset in cli with some tools? Or I have must use Afterburber too?

[thaneb]

Afterburner has nothing to do here, it's only nvidia-smi command line, installed along (included inside) nvidia drivers although memory offset can be set through nvidia-smi, I highly recommend you to use gminer to set it up, it allows its watchdog to reset settings in case of errors and at dag generation, using nvidia-smi won't allow this.

[thaneb]

sample of mix use of nvidia-smi and gminer oc settings , my start.bat @echo off nvidia-smi -rgc nvidia-smi -i 0 --lock-gpu-clocks=900,900 nvidia-smi -i 1 --lock-gpu-clocks=900,900 nvidia-smi -i 2 --lock-gpu-clocks=1425,1425 nvidia-smi -i 3 --lock-gpu-clocks=1005,1005 nvidia-smi -i 4 --lock-gpu-clocks=1050,1050 // note that i do not set anything for my 5th gpu (RTX 3060) miner.exe --algo ethash --server pool.address -n pool.port --user your.wallet.or.pool.userbname --worker name.of.the.worker --pl 54 54 61 59 70 70 --cclock 0 0 0 0 0 -500 --mclock +1100 +1100 +1100 +760 +760 +1200 --p2state 0 --tfan 58 58 58 56 56 60 --tfan_min 48 48 55 10 35 55 --tfan_max 65 65 65 45 65 65 --templimit 80 80 80 80 80 80 --electricity_cost 0.14 --api port.to.use.for.monitoring -l D:\Miners-Logs\log.txt

+---+-------+----+-----+-----------+--------+----+-----+-----+-------------+ | ID GPU Temp Fan Speed Shares Core Mem Power Efficiency | +---+-------+----+-----+-----------+--------+----+-----+-----+-------------+ | 0 3070 57 C 48 % 62.39 MH/s 45/0/0 900 8100 112 W 557.07 KH/W | | 1 3070 57 C 49 % 62.39 MH/s 63/0/0 900 8100 117 W 533.26 KH/W | | 2 3060Ti 58 C 56 % 62.44 MH/s 47/0/0 1425 8100 122 W 511.82 KH/W | | 3 2070 54 C 10 % 43.73 MH/s 42/0/0 990 7760 133 W 328.76 KH/W | | 4 1660S 56 C 39 % 31.52 MH/s 25/0/0 1050 7760 80 W 393.99 KH/W | | 5 3060 60 C 64 % 49.85 MH/s 47/0/0 1117 8700 119 W 418.93 KH/W | | 312.32 MH/s 269/0/0 683 W 457.28 KH/W | +---+-------+----+-----+-----------+--------+----+-----+-----+-------------+ 00:30:58 Pool: eth Shares/Minute: 4.01 00:30:58 Uptime: 0d 01:07:04 Electricity: 0.761kWh $0.11

[thaneb]

another great benefit of the clock lock through nvidia-smi, the 1660ti in my laptop went from 67degrees at 80wh to 56degrees at 58wh with +0.5 MH/s !!

[wacher74]

Ohh gminer, my god. Today it has been enchanted or something. I have used it for months, today it says: "Anti-hacking system detected modification of the miner memory. If you using an antivirus, add miner to antivirus exceptions" I don't know what is its problem. Yes, tiny MH/s improvements are here too.

[sebek72]

Runs unstable for me. Had to switch back to original setup with afterburner. My 470.05 driver gets automatically reverted somehow after each crash/reboot.

[HunteHax]

Any settings to recommend for RTX 2060 non super (Gigabyte)

Currently achieving around 31.40 M/hs, with 125 watts and 67.C temp

[pdieppa]

--p2state 0

What is this command? I don't think I found it in the documentation.

[thaneb]

After few days of mining and testing speed settings, I can say that using a locked core clock can really improve power consumption and hashrate on Turing gpu, but with ampere you need to give more freedom to the gpu (my 3070,3060 and 3060ti need to start mining with a "high" (especially for 3060/3060ti) Core clock, and then it will slowly go down to get closer to the smaller core clock your gminer was reporting while just using the core offset - 500) From my experience and Turing cards I'm using, 1660s,ti and 2070, core clock to - 450 was the best balance, and it gave me a core clock lock at 1050 for the 1660x and 1005 for the 2070.

[typekx]

I have a problem i can't change

[wacher74]

start cmd.exe as administrator as error message says.

[typekx]

Thx it's better

[thaneb]

Thx it's better

It's incredible, you got exactly the same issue I got, 2 rtx 3070, with the 2nd using more power than the first, I am unable to keep the 1st at 8100 memory clock without getting 1 rejected share per 2/3 hours, which I don't want to see.. What model and brand are you using ? I had to give higher core clock level to it in order to try to get better result with 8100, but it seems to doesn't work, so for now, i keep it at 8050... these results are made with these settings nvidia-smi -i 0 --lock-gpu-clocks=900,1020 (3070) nvidia-smi -i 1 --lock-gpu-clocks=840,960 (3070) nvidia-smi -i 2 --lock-gpu-clocks=900,1455 (3060ti) nvidia-smi -i 3 --lock-gpu-clocks=1005,1005 (2070) nvidia-smi -i 4 --lock-gpu-clocks=1050,1050 (1660s) nvidia-smi -i 5 --lock-gpu-clocks=1125,1545 (3060) now it seems very stable since 12 hours, and give me 2.5Mh/s more with 37Wh less than using core offset (-500,-500,-500,-450,-450,-500), even if reported coreclock of the 3060&3060ti remains higher than with offset (and they report higher clock rate, but lower wh, which is something i still don't understand)

[typekx]

Thx it's better

It's incredible, you got exactly the same issue I got, 2 rtx 3070, with the 2nd using more power than the first, I am unable to keep the 1st at 8100 memory clock without getting 1 rejected share per 2/3 hours, which I don't want to see.. What model and brand are you using ? I had to give higher core clock level to it in order to try to get better result with 8100, but it seems to doesn't work, so for now, i keep it at 8050... these results are made with these settings nvidia-smi -i 0 --lock-gpu-clocks=900,1020 (3070) nvidia-smi -i 1 --lock-gpu-clocks=840,960 (3070) nvidia-smi -i 2 --lock-gpu-clocks=900,1455 (3060ti) nvidia-smi -i 3 --lock-gpu-clocks=1005,1005 (2070) nvidia-smi -i 4 --lock-gpu-clocks=1050,1050 (1660s) nvidia-smi -i 5 --lock-gpu-clocks=1125,1545 (3060) now it seems very stable since 12 hours, and give me 2.5Mh/s more with 37Wh less than using core offset (-500,-500,-500,-450,-450,-500), even if reported coreclock of the 3060&3060ti remains higher than with offset (and they report higher clock rate, but lower wh, which is something i still don't understand)

Gigabyte GeForce RTX 3070 AORUS MASTER 8GB GDDR6

[thaneb]

wow, I am too much surprised that you can't get the 62+Mh/s with your cards, 3 fan and so a large cooler, should be easily reachable (my zotac are small sized with only 2 fans !) give them a try at PL 54, with coreclock at 960

[typekx]

I can on 900 core i have 62.63

[thaneb]

these cards have a general termal limit at 80°c, so as long as you remains under 62-64°c, you are very cool with them!

[typekx]

But have bad opinion about temp on vram. What config U propose

[thaneb]

from what you screenshot is saying, it's you current core clock which let you under the 62Mh/s, vram temp is related to your 8100Mhz, if you are mining like that without any rejected shares (which is what I supposed when seeing the size of your cards and of their cooling system), and when you see many miners going at 8200Mhz or more, I would give a try same I told you, with a nvidia-smi -i 0 --lock-gpu-clocks=840,960 nvidia-smi -i 1 --lock-gpu-clocks=840,960 PL 54, and keep unchanged the vram speed

[typekx]

from what you screenshot is saying, it's you current core clock which let you under the 62Mh/s, vram temp is related to your 8100Mhz, if you are mining like that without any rejected shares (which is what I supposed when seeing the size of your cards and of their cooling system), and when you see many miners going at 8200Mhz or more, I would give a try same I told you, with a nvidia-smi -i 0 --lock-gpu-clocks=840,960 nvidia-smi -i 1 --lock-gpu-clocks=840,960 PL 54, and keep unchanged the vram speed

Have You any proposition for 3080?

[gminadak]

@HunteHax

Any settings to recommend for RTX 2060 non super (Gigabyte)

Currently achieving around 31.40 M/hs, with 125 watts and 67.C temp

Try this Run your .bat as Administrator

--cclock -260 --mclock 855 --lock_voltage 700 --tfan 45 --tfan_min 60 --tfan_max 95 --templimit 60

2060 46 C 79 % 32.13 MH/s 14/0/0 1005 7655 80 W

With chrome open and room temp 28C if i close chrome etc i have 32.55 MH/s

MSI GEFORCE RTX 2060 VENTUS 6G OC

[HunteHax]

@HunteHax

Any settings to recommend for RTX 2060 non super (Gigabyte) Currently achieving around 31.40 M/hs, with 125 watts and 67.C temp

Try this Run your .bat as Administrator

--cclock -260 --mclock 855 --lock_voltage 700 --tfan 45 --tfan_min 60 --tfan_max 95 --templimit 60

2060 46 C 79 % 32.13 MH/s 14/0/0 1005 7655 80 W

With chrome open and room temp 28C if i close chrome etc i have 32.55 MH/s

MSI GEFORCE RTX 2060 VENTUS 6G OC

Just saw someone had replied. Will try it tomorrow Thanks in advance

[HunteHax]

@gminadak your settings for the 2060 were perfect Wanted to ask if you have any for 1060 that I could try ?

[yigitmengc]

I am using 2060Super. effective speed is constantly falling and I am losing share. I would be very happy if someone gave the oc settings 🥰

[ryanhuguelet]

1660 Super (bios mod) 34.5 mh/s pl75 --lock_cclock 1000 --mclock 950 2060 max Q 27.8 mh/s pl 60 --lock _cclock 1000 --mclock 1325 3060m 38.5 mh/s pl 60 --lock_cclock 1000 --mclock 699

[gminadak]

@HunteHax I don't own any 1060, sorry for the late reply but i didn't get notification. Well since i had participate in this thread and i got a notification i will post my current settings. First off all I'm using HiveOS only now and my settings have been changed a bit. And my hashrate now is 32.90 not big difference.

[HunteHax]

Well, after months spent in searching, what I got at best was 24 Mh/s.