ethereum-mining / ethminer

Ethereum miner with OpenCL, CUDA and stratum support
GNU General Public License v3.0
5.98k stars 2.29k forks source link

GTX 1060 6GB - max. Mhs 20.66 #119

Closed highree closed 6 years ago

highree commented 7 years ago

Hi,

is it just the real limit of the card? Whatever i do under windows 10 with and without updates i get max 20.66Mhs. Any thoughts?

blinkin69 commented 7 years ago

u use the latest version? https://github.com/ethereum-mining/ethminer/releases/tag/v0.11.0

you should get 26-28 with -cuda-parallel-hash command line option.

highree commented 7 years ago

yes! Latest. I waited for it. But i think its some driver - OC related stuff. I thought mabe someone has same behavior. The thing is whatever i overclock which settings which drivers no more than 20.66. Even i use different vendors: Gigabyte, EVGA, Palit, Gainward. for all the same. Im gonna give debian a try.

i also tried --cuda-parallel-hash 1 2 4 6 8

the best is default 4

blinkin69 commented 7 years ago

ok, i dont know for windows, but try --list-devices List and check if it is cuda or opengl, you use -U also right? i had similiar issues because i was on opengl driver and not with cuda.

highree commented 7 years ago

--list-devices reports me cl and cuda devices. yes i am using the -U option. ethminer reports cuda mining tough. Which linux distri do you use? Ive read that ubuntu 16.04 should be the best, can you confirm this?

jimmykl commented 7 years ago

How much are you overclocking them? I think 1060s need a decent memory overclock to get to mid-20s with Ethminer. Also check the RAM on the cards with GPU-Z – Samsung is the best, Micron is ok, Hynix not so good.

highree commented 7 years ago

I am clocking Memory +800 and GPU +100 (at the moment 0 - cuz it makes no difference). Powerlimit is 65% (also here it makes no difference if i push it up to 110%). Alltogether it uses 90w each.
Memory is Samsung.

Thank you

jimmykl commented 7 years ago

I'd try upping the power limit. 65% is aggressively low and may be forcing the cards to throttle down. Do some testing working down from 100% and you should find a point where the performance drops off – then keep them just above that.

highree commented 7 years ago

yeah, i tried that allready. The last 4 days to be exactly. my last hope was the release today.

jimmykl commented 7 years ago

i guess you've probably already tried the windows creator's update and the latest nvidia drivers?

highree commented 7 years ago

huh? windows creator's update is not released yet, isnt it ?

jimmykl commented 7 years ago

https://blogs.windows.com/windowsexperience/2017/04/11/how-to-get-the-windows-10-creators-update/

highree commented 7 years ago

started the download. Keep you informed about what happens... thx for now

highree commented 7 years ago

so ive tested with creators update, same results. Parallel i installed ubuntu and clocked the card with coolbits 31: GPU +100, MEM: +800 - no power limit: 19,80-20,24Mhs MAX.

Any thoughts?

(trying debian now instead of ubunto 16.04)

Felix4567 commented 7 years ago

Same problem here, I have 6x MSI GTX 1060 Gaming X+ 6GB and I get 121-125 Mhs for the rig :( overclocked with MSI afterburner +800 mem. The cards have all Samsung memory.

highree commented 7 years ago

@Felix4567 what os do you use ?

Felix4567 commented 7 years ago

@highree Also Windows 10, latest updates.

highree commented 7 years ago

seems that you also hit the 20mhs limit (if there is one)

blinkin69 commented 7 years ago

and with only one card the same? find out which RAM you have on cards. seems to be hardware limitation. try also one card on an other computer if possible...

pls have a look at https://github.com/ethereum-mining/ethminer/pull/18

derubm commented 7 years ago

i am using +950 memory, +250 gpu, as nvidia tries to clock down the gpu on higher memory, at least with my 1060 mini 6gb, getting constantly ~24mh / second, it results in the end at following speeds/limits: power limit 62% fan speed ~70 % (manual fan curve) GPU: ~1600 mhz, Memory: 9500 ( Samsung ddr5 ) miner is the latest version , measured wattage for one card is about 70 watts. image

edit: OS is WIndows 10 , make sure you have samsung memory, tune memory in 50 + steps, watch miner. also it could be that i am only lucky on the ram charge zotac used with my card. +950 seems pretty much the upper limit.

highree commented 7 years ago

i would be happy with 24mhs - could't figure out yet. What os do you use ?

JayChickenK commented 7 years ago

win 10 has a problem with p0 states on nvidia cards.

  1. Close your mining program.
  2. Download https://ufile.io/gi122
  3. extract it
  4. execute the nvidiasetp0state.exe
  5. start your miner again

on my win 10 gtx 970, from 16 --> 19 mh/s PS: no virus it is the nvidiasetp0state.exe from nicehash miner, but without the miner. if u dont trust me (can understand that) download nicehash miner and in the folder execute the nvidiasetp0state.exe

ghost commented 7 years ago

Is the activation of p0 state maybe only something for GTX 900 series and not needed on the 1000 series cards?

highree commented 7 years ago

i would not recommend to use such software. i think its kinda riskware!

ghost commented 7 years ago

@chfast Can we add support to go into p0 state into the miner itself if it helps for specific cards? (I did not test it)

I also don't trust any random file and hope every user here knows that any downloaded file from any github user could also contain a virus / ransomware. :-)

jimmykl commented 7 years ago

The original source of that exe is Nicehash.

If you want to be sure you're running a valid version download their latest build https://github.com/nicehash/NiceHashMiner/releases unzip and nvidiasetp0state.exe is in there.

Instead of running that exe you can set it to be on/off using nvidiaProfileInspector. You're looking to turn off 'CUDA - force P2 state' in Section 5. I got this from https://github.com/nicehash/NiceHashMiner/issues/1000

@JHGitty you're right it doesn't work on 1000 series cards. It did give my 970 a little boost.

ghost commented 7 years ago

Ok - so what's about integration of this performance boost on our miner here? :-) Of course only on "supported cards" (as I understand the whole GTX 900 series)

jimmykl commented 7 years ago

I think it's best to leave this at the profile level rather than trying to manipulate it from within the miner at this stage. Eventually maybe but personally I'd prefer to see dev time spent on better error handling or monitoring etc. before a slight boost for 900 cards which can already be achieved with some manual configuration :-)

ghost commented 7 years ago

You are fully right! 👍

chfast commented 7 years ago

Personally, I don't have an idea how to do it from code. Is this supported by CUDA API? If yes, I'm for adding another command line option to go into specified P-state.

ozdemiryucel commented 7 years ago

New gtx1060s have this problem, so 9 Gbps editions. I don't know why, I research it but I could not find its reason yet. I have asus gtx1060 9gbps edition. I realise that when old gtx1060s work in mining, they have %100 Memory Controller Load in GPU-Z, but new gtx1060s have maximum %91 Memory Controller Load. I think this is a clue. http://cryptomining-blog.com/tag/gtx-1060-9gbps-hashrate/

highree commented 7 years ago

@Guhong01 actually my memory is 100% under windows 10.

ozdemiryucel commented 7 years ago

@highree Are you sure? This is just for mining as

highree commented 7 years ago

yes i am dude.

ive got it work by 25Mhs Now. Tweaking a bit and tell you more.

jimmykl commented 7 years ago

@highree nice! do you know what worked to get it past 20.66Mh/s?

ozdemiryucel commented 7 years ago

@highree I will appreciate that. What is model of yours?

Felix4567 commented 7 years ago

My memory controller load is also at 88% - 90% max on all the cards.

@highree can you tell us your solution?

highree commented 7 years ago

@All yes i got 25Mhs now. Since it doesent work on linux windows etc with nvidia i tried nvidiainspector 1.9.7.8 under windows 10. Seems that clocking with this tool is pain in the ass cuz u need to clock each card by hand.

The results so far:

EVGA 1060GTX (GPU: +100, MEM: +800, PS: 0) ~25Mhs ZOTAC/PCPARTNER 1060GTX (GPU: +100, MEM: +800, PS: 0) ~25Mhs

Testing more cards now! More 2 come

ozdemiryucel commented 7 years ago

@highree you use 8 gbps edition. 8 gbps editions have not any problem about that, actual problem on 9 gpbs edition as mine and @Felix4567

highree commented 7 years ago

no i have 8 gbps version too. Seems that i can get these rates only on Zotac/EVGA. My Gigabyte rigs seems to stay at the limit -.-

The worse thing is that most of my cards are gigabyte.

ozdemiryucel commented 7 years ago

@highree :D I said so too, you use 8 gbps edition :D

highree commented 7 years ago

sorry, i meant to say that Gigabyte and Zotac/evga are 8gbps. But gigabyte is not pushable. What i see now is that the 25 are not stable. Not a full hour and it drops to 23.

The P state seems to be buggy somehow.

ozdemiryucel commented 7 years ago

@Felix4567 I reasearch still this problem. If you find solution, can you inform me?

Felix4567 commented 7 years ago

@Guhong01 Yes I will do. I'm currently looking for a solution. Trying different drivers/os and settings.

blinkin69 commented 7 years ago

@highree i have also gigabyte but gtx 1070 and have push it to with only +1200memclock @ 32 each (im just tweaking)

so i dont think its gigabyte.

highree commented 7 years ago

@blinkin69 i have 1060GTX would love to have some 1070 :D

blinkin69 commented 7 years ago

next week i get palit and someother 1060 and i will test and reply.

marcovillar commented 7 years ago

I have private version of ethminer reach to 27-28 mh/s on 1060gtx. other settings GPU -200 MEM +790

defabricated commented 7 years ago

@marcovillar Which GTX 1060 model do you use? This one with 9gbps memory?

marcovillar commented 7 years ago

basically it will increase any 1060gtx card by 2 - 3 MH/s but i am using the chinese version of the p106 card (cmp-6-1)

highree commented 7 years ago

@marcovillar would you share your private version with us ?