I’ve been using underclocking method of making the voltage/clockspeed curve in Afterburner completely flat. It gives you slightly better efficiency than just the regular -502 offset, although not really by a lot.
Here is an old comment that links to a video how to do it, if you are curious.
Now about a week ago I adjusted my cards so I have 3×3070 on one machine. I’ve been thinking of doing some comparisons with them since with same settings they should all perform about the same. Previously I’ve had some of them in same machine and some of them on my main computer so it’s not been possible to do good comparions.
I noticed however they are not quite performing exactly the same. All of them had about 1000 MHz core clock, and +1400 MHz on memory. Pool difficulty is 4G.
GPU #0: NVIDIA RTX 3070 – 63.02 MH/s, [T:44C, P:117W, F:69%, E:539kH/W], 7096/7102 R:0.08%
GPU #1: NVIDIA RTX 3070 – 63.02 MH/s, [T:53C, P:117W, F:85%, E:539kH/W], 7147/7147 R:0%
GPU #2: NVIDIA RTX 3070 – 63.01 MH/s, [T:46C, P:111W, F:84%, E:568kH/W], 7077/7096 R:0.27%
Hashrate: 189.06 MH/s, Shares/min: 3.008 (Avr. 2.775), Avr.P: 345W, Avr.E: 548kH/W
WD: 5 days 7 hours 52 mins 38 secs, shares: 21320/21345 R:0.12%, restarts 1
You immediately notice few things. The first one is clearly coolest. This is partially because it’s the only 3 fan. The first and third perform about the same. But oddly, the second one was performing about 1% better than others. And with 7000 shares, you would think it’s statistically accurate.
The first one is oldest. It’s been mining over 6 months now, and prior to that I had it about 2 months while being blissfully unaware of mining. Oh how I would have a huge load of 3060 Ti’s if I would have known how profitable this was in December when it was relatively easy to get them. The third one I’ve had quite long time too, maybe from February, but it has mined at least one month less than the first one. The second one is newest, that has been mining nearly two months now.
But, with that way of underclocking, each GPU will automatically adjust it’s core voltage to the smallest it can fully work. The first and last one were operating at 716/718 mV, while the second one was one notch higher at 725 mV. So I didn’t really think the second card is performing better because it’s newer, I thought it’s because of it’s higher core voltage. So I decided to increase each of them by one notch, and reset the miner.
When I initially looked the situation about 17 hours later, they had mined all something around 605 shares. The numbers were very similiar so I thought maybe that was it. But sadly no, now after 35 hours later after I reset them
GPU #0: NVIDIA RTX 3070 – 63.03 MH/s, [T:46C, P:121W, F:69%, E:521kH/W], 1981/1983 R:0.1%
GPU #1: NVIDIA RTX 3070 – 63.04 MH/s, [T:56C, P:119W, F:85%, E:534kH/W], 1971/1971 R:0%
GPU #2: NVIDIA RTX 3070 – 63.05 MH/s, [T:48C, P:116W, F:84%, E:548kH/W], 1931/1933 R:0.1%
Hashrate: 189.12 MH/s, Shares/min: 3.874 (Avr. 2.816), Avr.P: 354W, Avr.E: 534kH/W
WD: 1 day 10 hours 50 mins 1 sec, shares: 5883/5887 R:0.07%
So now the first and second are performing about the same, while the third one is still lagging behind. But perhaps it’s not long enough time for them to avarage it out. I have been hoping I could get them close enough, so I could compare bunch of miners and bunch of pools. But it does seem it improved the overall hashrate. I just don’t know if I can make meaningful comparisons if different cards can have 1% – 2.5% differences.
I think I will give it a weekend and see if they get close enough to each other, to do more intresting tests while having a reasonable accuracy.