Getting more than 30 hash ETH on GTX 1070

RevMen

Limp Gawd
Joined
Feb 22, 2004
Messages
460
How are you folks getting 31-32 MH/s on your GTX 1070s?

I have 3 different models and I haven't been able to do better than 29 MH/s on some cards. The biggest disappointment is some MSI Aero's that won't go over 25 without going up to or over 90C.

What OS, drivers, miners are you using? And what settings do you have for your cards?
 
I have 2 1070s an EVGA Superclocked and an ASUS ROG Strix, both get around 29-30 by bumping up the memory clock by +500 or so. Any more than that and I get errors and screen artifacts.
 
I have an MSI Armor 1070, +600 memory 65% power limit -50 core clock which hit 31.2, but I ended up dropping it down to +500 memory because I started getting massive artifacts while browsing + mining. Latest drivers. ClaymoreEth on Ethermine.
Whats the temperature/airflow in your room? I also (strangely) found that stopping/restarting my miner would change my reported hash by up to 1Mh/s.
Also my effective rate has been better than my reported for as long as I've had the 1070.
 
If your card is going over 90C then you're doing something wrong. Or the card is broken.

Now, I don't own a 1070, I have a 1080Ti and 10x1060. on 1080Ti I use +125 gpu and +800 memory and it works fine, 1060 +125 gpu and +600 memory.
Do not set TDP to above 80%, really, it's not a big deal if your cards GPU is clocked down, for ETH it's memory clock that matters, not GPU, and lowering TDP only lowers GPU clocks, not memory.
 
Last edited:
If your card is going over 90C then you're doing something wrong

Clearly. Hence this thread.

I don't think it's a broken card thing. I've got 4 of these cards and they're all struggling in the same way. On the same rig are 2 EVGA FTWs and they run cool as cucumbers.
 
My point being, the card should not be up that high, at all. Mining Ethereum puts less strain on the GPU than games. You can lower TDP all the way to 60% or less, which means it's spending half the power it normally does. The temps should be really low!
For example, my 10 1060 are at about 60C when under 55% TDP, but 82 when under 100% TPD.
You don't need high TDP to mine Ethereum. Bump memory as high as you can without getting artifacts, check your hashing rate at 100% TPD, then lower TDP to 60%, check your hash rate again. It will be a bit lower, but is it acceptable? For my 1060 cards going from 85% TDP to 55% TDP lowered hash rate from 94 > 92mh/s and temps from 73C to 60-62C and power draw from 500W to 375W. Totally acceptable.
Once you do that and the card is still 90C, then it's either broken or you're sitting on it. Check your cooling. Are they on PCI-E risers? If yes, spread them out a bit, if you've stuck all 4 cards next to each other on the motherboard... then why?
 
Clearly. Hence this thread.

I don't think it's a broken card thing. I've got 4 of these cards and they're all struggling in the same way. On the same rig are 2 EVGA FTWs and they run cool as cucumbers.
How do you have them mounted? I have a pair of Asus 1070s, and if they're installed right next to each other, the top one hits 83 C almost immediately, and then throttles back. Installing them with a couple of slots' worth of gap between them makes a huge difference. The Founder's Edition cooler is a little less susceptible to this than a lot of partner designs are, even if it's less effective overall.

As Nebell said, your cards actually may have something wrong with them if they're hitting 90C in a relatively cool environment. All the Pascal cards I've seen try REALLY hard to not go above 83. Maybe see if you can remove the cooler and redo the thermal grease?

I'm with you though, in that my 1070s don't seem to go faster 27ish MH/s, regardless of how I adjust the overclock settings.
 
Its quite normal for the top card of a set of GPUS stacked right on top of each other to be substantially hotter (10-15C) than the bottom one. Hot air rises and the top card will be dealing with not only its own heat but the heat given off by the bottom card. As others have mentioned, a bigger gap or risers would help. If the cards are already spaced nice and far apart then there is something wrong with the card.
 
I own a couple of MSI gaming X 1070 8Gb with Micron memory. I have the fan set to 70%, memory to +800, power to 90, and it cranks out 32-34 MH/s at 60c and 75% TDP? Would have bought a couple more except I bought all that I could find on Okinawa.

Here hoping I find one in Osaka or Nagoya.

Drivers are 374.xx.
 
Maybe you guys have bad memory? 1060 should has about 19 out of the box and 23-24 oced. All my 10 1060 hash 15 out of the box and 18.5-19 oced. It's because they have a shitty Xynix memory. There is no way around that. It's just bad luck.
 
I own a couple of MSI gaming X 1070 8Gb with Micron memory. I have the fan set to 70%, memory to +800, power to 90, and it cranks out 32-34 MH/s at 60c and 75% TDP? Would have bought a couple more except I bought all that I could find on Okinawa.

Here hoping I find one in Osaka or Nagoya.

Drivers are 374.xx.

What OS and what miner are you using?
 
Here's the rig. Seems like a decent amount of space between cards, but I could be wrong.
IMG_20170622_100341.jpg

IMG_20170622_100352.jpg
 
So I'm guessing I'm doing something wrong if I'm getting like 2.8 - 3 MH/s with a 980Ti?
 
My EVGA SC ACX 3.0 cards were able to hit 31 while running pretty cool, and they were "1 slot spacing" apart, meaning two actual motherboard PCIe slots apart. The top card ran a bit hotter than the bottom because it was sucking hot air in, but it wasn't bad. My Antec 300 case has a fan that blows air right onto the PCIe slots, so there was some fresh air getting in. Wish I could remember what the settings were I was using.. at the time I was mostly mining ZEC. They were never over 75C though.

RAM was at 9200MHz, I remember that at least.
 
Here's the rig. Seems like a decent amount of space between cards, but I could be wrong.

Take the shrouds and fans off the cards, build a cardboard tunnel around everything and direct airflow down with a simple box fan. Why?

1) The cards, PSU and mobo will run cooler - more OC at less volts - profit.
2) You won't be replacing dead fans that aren't meant to run 24/7 - less downtime, profit.
3) when you sell the cards, the fans and shrouds will look brand new... because they will be - profit.
4) a box fan is ten bucks at Walmart and moves more air much more quietly than all the best PC fans fans - hearing is spared, less money on hearing aid batteries - profit.
 
The box fan idea is good. Thanks.

I moved the power limit down to 60% and now my temps are low 80's. So big improvement there.
 
This is what I'm getting on Ubuntu with 866 watts from the wall with 6x 1070s dualing mining ETH/SC.

ETH - Total Speed: 180.068 Mh/s, Total Shares: 2000, Rejected: 0, Time: 15:48
ETH: GPU0 30.198 Mh/s, GPU1 30.240 Mh/s, GPU2 30.186 Mh/s, GPU3 29.500 Mh/s, GPU4 30.086 Mh/s, GPU5 29.857 Mh/s
SC - Total Speed: 1200.451 Mh/s, Total Shares: 429, Rejected: 0
SC: GPU0 201.322 Mh/s, GPU1 201.599 Mh/s, GPU2 201.240 Mh/s, GPU3 196.667 Mh/s, GPU4 200.575 Mh/s, GPU5 199.048 Mh/s
GPU0 t=62C fan=75%, GPU1 t=48C fan=75%, GPU2 t=64C fan=75%, GPU3 t=61C fan=75%, GPU4 t=60C fan=75%, GPU5 t=49C fan=75%
 
I am using nvOC for Nvidia GPU mining on Ubuntu.
Link here

As for the settings, I have the following settings:
5 cards:
Overclock
cc 100
mc 1275
Powerlimit: 125 watts
1 card:
cc -100
mc 1225
Powerlimit: 120 watts
(this card does not like OC had to tweak it some for it to not cause instability within the rig)
 
For me, the biggest thing is having memory running @ 9000mhz (2,250mhz x4 multi for GDDR5) in MSI afterburner.

That seems to keep me right at 30MH

also, I underclock the core by setting the TDP/power setting to 70%, that brings the core clock down a few hundred mhz and doesn't affect the speed.

also, my whole rig is only pulling 190watts from the wall which is pretty sweet too.
 
Last edited:
I got my 6x 1060 rig up today. Out of box each one was around 18 Mh/s, but I did an aggressive OC on the memory to +1500MHz and now I'm getting 22 Mh/s solid. It's only been running for a few hours but it seems alright. Will let it run overnight and see if the OC holds.
 
So is one card better than another? Just started mining and my evga 1070 sc black ed gets 31 MH/s with small o/c

I have an Asus Strix OC on the way. Wonder if it's any better or should I just stick with evga?
 
Just started ethereum mining using genoil, NVidia inspector to do the ocing..... EVGA GeForce GTX 1070 SC GAMING, 08G-P4-6173-KB, 8GB GDDR5, ACX 3.0 & LED..... micron memory, I can get memory to +600-640 stable, with 65% power.... ive messed around with 700, seems stable for a bit then kinda wonks out.... 800 same thing... got core at -200, pulls around 29-31MH/s 61c under load

As a side note, I did downgrade my GPU Bios firmware....think it was 86.04.50.00.72..... heard that 85.04.50.00.70 was better for overclocking, can confirm I went from +500 on memory to +640....as far as core, befor I was mining, I had it running around +200 but saw neglidable performance gains tbh...and a lot more heat for my Azza Mini-ITX case...
 
mining is very memory intensive you want the best performance get either founders edition models or (not too sure about these) SSC or above. All nvidia founders editions are samsung memory. They overclock to 1ghz pretty much guaranteed. Actually any card with with 8000 mhz samsung. If you have micron you are SOL. They are shit at overclocking and you would be lucky to get 500+ OC on those. Samsung GDDR5 is the best. Core clock does not do anything, dont even mess with it. If you are using riser cards don't lower the power too much. Leaving at 75-80% power your cards will barely get 70% there is just that loss there. This is good for long term stability.

I am getting 30-31 on all my founders edition cards using about 70% power average. As an example my 1000w gold powe supply is pulling 700w under load with 6 gtx 1070s.

As an example. I have 3 1060s all evga, Two are regular gaming, micron memory. One is SSC samsung memory. My Gaming card with micron memory OCs about 700mhz on memory while SSC is rock solid at 1ghz. Difference between samsung and micron. This is the reason I decided to go with founders edition cards so I can have consistent results all across for 1070s.
 
Ive been doing a lil more tweaking, but yeah, micron memory is just not as good as the Samsung... but heres my results for the last 45 minutes
 

Attachments

  • current.png
    current.png
    718.3 KB · Views: 548
Just fired up an MSI Gaming X 1070 today. Set PL to 50% and memory to +700. Hashing away a little under 31Mh/s all day. High temp of 64C, and that's in my garage where it is probably 40-45C from the ~6kw of other machines in there.

EDIT: Slightly under 31. Over 31 must have been before I dropped the power limit.

EDIT: Micron memory. Not actually sure how fast it is running. Afterburner shows stock is 3800 and +700 puts it at 4498. GPU-Z reads 2352MHz on the first tab and as 2249MHz on the Sensors tab.
 
Last edited:
Just fired up an MSI Gaming X 1070 today. Set PL to 50% and memory to +700. Hashing away a little under 31Mh/s all day. High temp of 64C, and that's in my garage where it is probably 40-45C from the ~6kw of other machines in there.

EDIT: Slightly under 31. Over 31 must have been before I dropped the power limit.

EDIT: Micron memory. Not actually sure how fast it is running. Afterburner shows stock is 3800 and +700 puts it at 4498. GPU-Z reads 2352MHz on the first tab and as 2249MHz on the Sensors tab.

yea a bug in reporting in eiter driver or afterburner. Be on the lookout. If memory is unstable you will get either crash, the miner hang with a cursor or computer will hard lock. It may run for a day or two and then you notice an issue.

out of 3 micro based cards I didn't have one that could sustain 1ghz over long period they all went back to best buy and ordered founders edition.
 
Just got a PNY FE running too, though interestingly dropping the PL to 50% drops the core to 800MHz, and it doesn't hash more than 21Mh/s. Had to bump PL to 70% to get the card's clocks back up into the low 1500 range, and it is churning out low 30.2-30.3.

I wonder if the MSI Gaming X has a higher TDP than it advertises. Their site says 150, but maybe it is coded closer to 200w or something since it has 8+6 power? I don't know.

I know when I had my EVGA cards, I played around with the voltage/MHz curve to get things running faster at lower voltages. I don't really feel like screwing around with that for these cards though.
 
so have all the 1070 manufacturers switched to micron memory? or do some still use samsung memory?
 
THe PNY FE card I mentioned above is using Samsung. Mind you, both of these cards were purchased used recently, so I don't know how old they actually are.
 
so have all the 1070 manufacturers switched to micron memory? or do some still use samsung memory?
I have an Asus 1070 (the "dual" model, as I recall) that has Hynix memory.

It's installed in the same system as a Strix model that has the Micron memory, and they're both pretty weaksauce overclockers memory wise. No way I'm getting 800+ MHz on the memory like some folks are claiming they're getting.
 
Did some searching last night, heres the page I got the info about bios updates from for evga with micron memory...

https://forums.evga.com/EVGA-GeForce-GTX-1070-BIOS-Update-v8604500070-m2565056.aspx

hope it helps some of you out there with your oc's and hashrates..... I seriously think I can get my card to 800, assuming I wanted to do more then just flash the bios...... 700 stable, 750...ish...for 30 minutes...... 800... get about 5 minutes......

the bios mod was as far as I'm willing to go...so 700 it is, with 31~ MH/s

EDIT: looking through that links many pages of post, it seems the 86.04.50.00.72 bios I had supposedly included this fix, however I know for a fact befor I was only able to get +500 MAX...which realistic probably equated to 300 or so on the memory......FWIW
 
Last edited:
I'm getting 30MH/s on my GTX 1070, and I confirmed earlier today that it has Micron memory.

It has been running stable @ 9000mhz memory for a few days now.
 
Also got a Zotac 3-fan one with Micron memory up and running at 70% and +700 memory hasing in the mid 30's.

Using Windows 10 and the 381.89 drivers. I had installed the latest drivers with the first card and had some weird speed issues, so I downgraded to the drivers that I have currently installed on my main desktop with the 1080 Ti. Turns out the speed issue was a bad riser, but I haven't bothered upgrading the drivers to current since things seem good where they're at.
 
770 + 1070 hitting 45-46Mh/s total, versus 38-40 on Claymore Miner at same clocks.

Ethermine website stats had me at a solid average at 40Mh/s, so I'll check back in a 24hours or so and see if the 45Mh/s reported actually produces.
 
ugh!

Think I just made a mistake. Had 2x EVGA 1070 SC black ed on order from amazon.

I canceled it b/c I snagged 2x EVGA FTW Gaming versions for the same price. I was looking for ones with a higher chance of getting Samsung memory. Now, I find out that the FTW draws more power and it's likely Micron memory. Looks like 1060 6gb are the best in terms of cost/hash rate ratio.
 
I'm not 100% on this, but I'm pretty sure the SC versions were micron, FTW was Samsung.... id stick with what your ordered.....
 
Back
Top