What speed you can manage on for example 650mhz on which most cards work?
Using newest ccminer that has equihash, compiled optimized, cuda 9.0, I'm getting ...
Solo gtx-1070 I run at +1600, above that it crashes
All cards about +150 over clock
When I mention 1600 I'm talking transfer
Gtx-1070 in mining rig usually +1200, but I run the -pl ( power watts ) at below 120, I like 110
Gtx-1060-6gb likes +1000 on memory-transfer
Gtx-1060-3gb seems to not like over +500 on memory transfer
[ 1070's in rigs I find prefer not over +1000; When running EWBF I find it crashes to often if I run at +1600; I don't find any crash problems in the new ccminer, solid as a rock and I love having the software so I can fuck around with the .CU files and optimize my code ]
On all cases I get
500 sols/sec on 1070
300 sols/sec on 1060's ( -3 or -6 gb almost the same, maybe -6 5% better, but most likely because I run -3 at 85 watts and -6 at 95 watts )
3 sols/watt 1060
4 sols/watt 1070
I keep my rigs (cards) below 80c , my 1070s run now at about 60c, and my 1060's run at 68c auto-fan, but I also have rig fans in front ( std 120mm fans blowing in ), and I have large 80cm cross fans running through racks to move hot air
More important is COST
My cost for 1060-3 is $200 USD, 1060-6 is $300, and cost of 1070 is $600, thus the best bang for BUCK is the 1060-3 same efficiency as the -6, but -30% the price, the 1070 in comparsion sucks
The 1080 costs $1k USD where I live. I usually buy my 1060's from GALAX, which is straight from china. Good discount on 1060-3 nobody wants them, thanks to ethereum
For equihash we only need 650mb of memory so the 1060-3 is just fine, no problem like found on ETHEREUM which is NO longer profitable to mine.
IMHO best be now for equi is rigs of $200 cards that be six, so cost is $1200, cpu biostar is $50, and rig is $30, cpu/mem is about $50; six 1060-3 does about 1800 sols/sec, right now generates about $15/day in zencash.
My power bill is $60 per rig, per month so 4days a month the rig pays the power, then all else is mine,
I don't mind buying GPUS as I have been doing MACHINE-LEARNING & CUDA forever, thus I can always redeploy my stuff to research anytime I wish.
***
Talking about the 'watt' issue above, well six cards at 80watts that's 480 watts and power-supply mother board is 120 watts so I target 600 watts per rig and keep them cool. Much less than $60/month electricity cost.
***
more important here I not see addressed is how are you controlling card? I'm on linux, I hate windows
linux-cuda requires control from nvidia-settings & nvidia-smi
most problems I is people are running windows & just use the MSI gui and set all to max,
My rigs make NO noise, I can't even hear the fans
I think its best to BUY all the cards on the market dial them all in with the different mining algos and know what to deploy for what problem at hand,
I would have never bought 1060-3's if I hadn't taken this approach, but now that I actually know its limitations and power, I find its the best for equihash for cost/performance
When you talk about +650 over clock on 1070, I don't think it really matter's 1060-6 or 1070's once you go over +500 the gain is negligible, its best to dial back the power to minimal, and learn how high you can clock before the miner software crashes
Like I have said I find +1600 to be the magic number that crashes all NIVIDIA software, for many app's I can run +1550 just fine all day,
But if I run EWBF I will get hourly crashes, if I want crash free I have to go down to +1000
With the new ccminer, I have not studied it limits, but I can tell you this at +650 your running its at less than 1/2 its potential ( my gut feeling so far is that I can run ccminer at +1400 just fine )
All the above said if your running at 150 watts, and +1600 over-clock you will fry you card, I think big mistake I see in these forums is people on windows told to set fans to MAX, and Power, ... and all over-clock ally you are doing is burning up your graphics cards.