How to Overclock your GPU for Mining (NVIDIA GUIDE) Max profit, hashrate & efficiency on any coin - Sebs FinTech
video description
Date: 2022-04-19
Related videos
Comments and reviews: 10
CaedenV
Ok! So inspired by this video, it has been a few weeks of fiddling with stuff, and found a few things with the efficiency game, and numbers to back it up!
1) Proper cooling is absolutely key!
Before messing with my paste and pads I had to run fans super high to keep things cool, and I just couldn't hit the efficiencies that I was finding others could get online. After redoing paste and pads, both of my cards got a decent bump up in performance, and there was no thermal penalty on turning the fans down. So that means less power use on the GPU, but also less power use by the overall system because of the fans.
GTX1080 would max out at 23-24MH/s at 110W before running into stability issues, but after redoing the paste and pads it is getting 30-31 at 95W, and I was able to turn the fan down from 95% to 60%
RTX3090 would max out around 112MH/s at 310W, but never hit any stability problems. Instead it would just overheat and back itself off and loose performance. After redoing the pads and paste it is getting 118MH/s at 290W, and the fans can run at 65% and keep at 96-98c. It makes the basement much more bearable to be in lol.
But my point here is that reported temps are not the only temps that matter. On my 3090 even with the same overall temps before and after the thermal paste things could run much faster on much lower power input, which means that something wasn't getting enough cooling and was causing a bottleneck. Watching the reported temps is important, but just always remember that any part in the power delivery chain can get pretty warm and loose efficiency. I think that adding thermal pads to all of the little bits and pieces in the chain helped with what I was fighting just as much as keeping the RAM cooler without needing 90+% fans all the time. On the 1080 the reported temps were all downright cool, and yet the stability issues cleared up after adding dabs of thermal paste to additional components on the board that would not otherwise make contact with the heatsink. It can make a huge difference, and cost $30, so the improvements will pay for themselves in 2 months.
2) Each card is wildly different
On the RTX 3090 it is all about having a high RAM OC, and then under-clocking the GPU and limiting power for efficiency. So for mine, it is +1000 RAM, -125 GPU, and 85% power limit.
On the GTX 1080 the GPU needs an OC to keep the RAM busy, but OCing the RAM would not increase performance no matter what I have tried. Ultimately it has a -25 RAM, +150 GPU, and a 50% power limit.
Interesting to note too are the RAM timing modifications for the 1080 card (Affects all GDDR5x cards). Changing timing options gives a 50% increase in Eth performance, raising the max from 20-25MH/s to 30-33MH/s. EthPill can do this, but NiceHash has some built-in options for it as well. All that to say to not assume that what works for one card will work for another. 2 of the same model may work similarly, but every generation of card is going to have their oddities.
3) If you are using NiceHash, re-run your benchmarks when you are done messing with settings
For one, it will help your card to properly choose the most effective coin to work on with your new settings. For another, you may find that a different vendor's algorithm will give you better results. For the longest time Excavator was the defacto miner for my rig, but after adjusting settings to have less GPU under-clock I found that I can get closer to my theoretical limit using NBMiner's Eth algorithm. I re-ran the benchmarks a few times with the old -400 GPU vs the new -125 GPU settings I am using, and something about NBMiner just works better with a little more GPU power, while Excavator runs best essentially with my GPU as far down as it will go.
So if you change algorithms, have a driver update, or change your cooling setup, or change your OC be sure to re-run your benchmarks to make sure you are using the miner that can best make use of your settings.
4) Efficiency matters a lot!
I mean... I am using a GTX1080 to mine with, so I am not opposed to using what you have available to you. But use what you have to it's fullest advantage!
The 1080 had been 86 cents per day, and using 21 cents per day in power use, for a profit of 65 cents per day. With better performance and efficiency that is now saving 4 cents per day in power use, while increasing output to $1.14 upping profits to 94 cents per day. That is a small number, but a massive 45% increase in profitability, or $8.70/mo extra profit compared to previous settings.
On the RTX 3090 the numbers are all larger, but the overall percentage increase less impressive.
Old settings were using 60 cents a day in power for $4.26 in Eth, or a $3.66/day profit. New settings are using 56 cents of power for $4.41 Eth, or $3.85 in profit. That is a respectable 4% improvement, or $5.70/mo in extra profit.
Now I am a super small-time guy, but you multiply that kind of savings across a rig with 8 cards in it, or a farm with multiple 8 card rigs, and even a small 4% increase in efficiency becomes every important.
Along those lines, I don't have a Kill-a-watt meter (yet), so it is hard to tell exactly, but I also upgraded my power supply on the rig with the 3090 in it. With my overall use, moving from an old 80+ to an 80+Gold PSU should also be saving me $80/year in power costs. The PSU was on sale for $85 for a nice brand-name 850W PSU, and it is essentially going to pay for itself in year 1, and have enough power that I can add a 2nd GPU to my main rig as soon as I can nab something on sale or used. But assuming prices remained the same (lol), that $80 in wasted electricity would be 5% of the card's expected $1400 in profits. Now obviously I am hoping for coin prices to rise before I sell, so it would be much less that 5% at the moment. But when Eth goes away I am expecting my main rig to drop from $3.85/day down to $2/day. So $80 of $730 would be an 11% hit to profits, and that sucks. I am not sure that the extreme prices on Platinum rated PSUs will ever pay for the added efficiency they bring compared to a gold rating (1.5% better for 2.5x the price? That is crazy talk!), so there are limits to sensibility in efficiency, but that low hanging fruit is important to grab while you can.
reply
Ok! So inspired by this video, it has been a few weeks of fiddling with stuff, and found a few things with the efficiency game, and numbers to back it up!
1) Proper cooling is absolutely key!
Before messing with my paste and pads I had to run fans super high to keep things cool, and I just couldn't hit the efficiencies that I was finding others could get online. After redoing paste and pads, both of my cards got a decent bump up in performance, and there was no thermal penalty on turning the fans down. So that means less power use on the GPU, but also less power use by the overall system because of the fans.
GTX1080 would max out at 23-24MH/s at 110W before running into stability issues, but after redoing the paste and pads it is getting 30-31 at 95W, and I was able to turn the fan down from 95% to 60%
RTX3090 would max out around 112MH/s at 310W, but never hit any stability problems. Instead it would just overheat and back itself off and loose performance. After redoing the pads and paste it is getting 118MH/s at 290W, and the fans can run at 65% and keep at 96-98c. It makes the basement much more bearable to be in lol.
But my point here is that reported temps are not the only temps that matter. On my 3090 even with the same overall temps before and after the thermal paste things could run much faster on much lower power input, which means that something wasn't getting enough cooling and was causing a bottleneck. Watching the reported temps is important, but just always remember that any part in the power delivery chain can get pretty warm and loose efficiency. I think that adding thermal pads to all of the little bits and pieces in the chain helped with what I was fighting just as much as keeping the RAM cooler without needing 90+% fans all the time. On the 1080 the reported temps were all downright cool, and yet the stability issues cleared up after adding dabs of thermal paste to additional components on the board that would not otherwise make contact with the heatsink. It can make a huge difference, and cost $30, so the improvements will pay for themselves in 2 months.
2) Each card is wildly different
On the RTX 3090 it is all about having a high RAM OC, and then under-clocking the GPU and limiting power for efficiency. So for mine, it is +1000 RAM, -125 GPU, and 85% power limit.
On the GTX 1080 the GPU needs an OC to keep the RAM busy, but OCing the RAM would not increase performance no matter what I have tried. Ultimately it has a -25 RAM, +150 GPU, and a 50% power limit.
Interesting to note too are the RAM timing modifications for the 1080 card (Affects all GDDR5x cards). Changing timing options gives a 50% increase in Eth performance, raising the max from 20-25MH/s to 30-33MH/s. EthPill can do this, but NiceHash has some built-in options for it as well. All that to say to not assume that what works for one card will work for another. 2 of the same model may work similarly, but every generation of card is going to have their oddities.
3) If you are using NiceHash, re-run your benchmarks when you are done messing with settings
For one, it will help your card to properly choose the most effective coin to work on with your new settings. For another, you may find that a different vendor's algorithm will give you better results. For the longest time Excavator was the defacto miner for my rig, but after adjusting settings to have less GPU under-clock I found that I can get closer to my theoretical limit using NBMiner's Eth algorithm. I re-ran the benchmarks a few times with the old -400 GPU vs the new -125 GPU settings I am using, and something about NBMiner just works better with a little more GPU power, while Excavator runs best essentially with my GPU as far down as it will go.
So if you change algorithms, have a driver update, or change your cooling setup, or change your OC be sure to re-run your benchmarks to make sure you are using the miner that can best make use of your settings.
4) Efficiency matters a lot!
I mean... I am using a GTX1080 to mine with, so I am not opposed to using what you have available to you. But use what you have to it's fullest advantage!
The 1080 had been 86 cents per day, and using 21 cents per day in power use, for a profit of 65 cents per day. With better performance and efficiency that is now saving 4 cents per day in power use, while increasing output to $1.14 upping profits to 94 cents per day. That is a small number, but a massive 45% increase in profitability, or $8.70/mo extra profit compared to previous settings.
On the RTX 3090 the numbers are all larger, but the overall percentage increase less impressive.
Old settings were using 60 cents a day in power for $4.26 in Eth, or a $3.66/day profit. New settings are using 56 cents of power for $4.41 Eth, or $3.85 in profit. That is a respectable 4% improvement, or $5.70/mo in extra profit.
Now I am a super small-time guy, but you multiply that kind of savings across a rig with 8 cards in it, or a farm with multiple 8 card rigs, and even a small 4% increase in efficiency becomes every important.
Along those lines, I don't have a Kill-a-watt meter (yet), so it is hard to tell exactly, but I also upgraded my power supply on the rig with the 3090 in it. With my overall use, moving from an old 80+ to an 80+Gold PSU should also be saving me $80/year in power costs. The PSU was on sale for $85 for a nice brand-name 850W PSU, and it is essentially going to pay for itself in year 1, and have enough power that I can add a 2nd GPU to my main rig as soon as I can nab something on sale or used. But assuming prices remained the same (lol), that $80 in wasted electricity would be 5% of the card's expected $1400 in profits. Now obviously I am hoping for coin prices to rise before I sell, so it would be much less that 5% at the moment. But when Eth goes away I am expecting my main rig to drop from $3.85/day down to $2/day. So $80 of $730 would be an 11% hit to profits, and that sucks. I am not sure that the extreme prices on Platinum rated PSUs will ever pay for the added efficiency they bring compared to a gold rating (1.5% better for 2.5x the price? That is crazy talk!), so there are limits to sensibility in efficiency, but that low hanging fruit is important to grab while you can.
reply
Lee
Some numbers:
25.81 MH/s - 244 W - 106Kh/W (Stock)
23.93 MH/s - 125 W - 193Kh/W (Power Limit)
30.80 MH/s - 270 W - 111Kh/W (Memory Overclock)
30.82 MH/s - 278 W - 111Kh/W (Memory + GPU Overclock)
30.56 MH/s - 153 W - 200Kh/W (All: Power Limit + Memory + GPU)
In %
MH/s Stock vs All: +18 %
W Stock vs All: -60 %
Kh/W Stock vs All: +89%
For the next time, maybe prepare a little sheet. My 2 cents. Peace.
reply
Some numbers:
25.81 MH/s - 244 W - 106Kh/W (Stock)
23.93 MH/s - 125 W - 193Kh/W (Power Limit)
30.80 MH/s - 270 W - 111Kh/W (Memory Overclock)
30.82 MH/s - 278 W - 111Kh/W (Memory + GPU Overclock)
30.56 MH/s - 153 W - 200Kh/W (All: Power Limit + Memory + GPU)
In %
MH/s Stock vs All: +18 %
W Stock vs All: -60 %
Kh/W Stock vs All: +89%
For the next time, maybe prepare a little sheet. My 2 cents. Peace.
reply
Joey
Question
I noticed while I was fine tuning my OC settings that my shares/min would fluctuate anywhere from 0.5 up to 8 shares. When I found my sweet spot on my 3080 ftw3 LHR mining RVN, my shares/min dropped to 0.8. Should we be paying attention to this information in trex miner or just go with the sweet spot stability. I have the card purring at 53MH/ 285W / 187khW
oc setting power75 , core +200, Mem +1850 Fan 100%.
reply
Question
I noticed while I was fine tuning my OC settings that my shares/min would fluctuate anywhere from 0.5 up to 8 shares. When I found my sweet spot on my 3080 ftw3 LHR mining RVN, my shares/min dropped to 0.8. Should we be paying attention to this information in trex miner or just go with the sweet spot stability. I have the card purring at 53MH/ 285W / 187khW
oc setting power75 , core +200, Mem +1850 Fan 100%.
reply
Trent
Thank you for this video. I didn't think the gains I got were possible. I had a 3070 on ETH doing 50Mh/s at 54% power. Now it's doing 63Mh/s at 60% power. I had not done any tuning of the clocks as I thought the gains were minimal.
reply
Thank you for this video. I didn't think the gains I got were possible. I had a 3070 on ETH doing 50Mh/s at 54% power. Now it's doing 63Mh/s at 60% power. I had not done any tuning of the clocks as I thought the gains were minimal.
reply
bell
my gpu (3070 ti) gets 55 to 54 MH/s with a mem clock of +400 and a core clock of +100 with a efficiency of 287kH/h and 185.5 W and a power limit of 60% is that good
reply
my gpu (3070 ti) gets 55 to 54 MH/s with a mem clock of +400 and a core clock of +100 with a efficiency of 287kH/h and 185.5 W and a power limit of 60% is that good
reply
Juicy_Tender_Steak
the best and most simple video for OC i have ever seen, should we keep the Fan speed at 100% after doing all the overclocks or put it back to Auto?
reply
the best and most simple video for OC i have ever seen, should we keep the Fan speed at 100% after doing all the overclocks or put it back to Auto?
reply
Joseph
Seb,
Been watching for about 2 years and I must say, whenever I need help, your videos come through clutch!! Thank you so much for what you do!!
reply
Seb,
Been watching for about 2 years and I must say, whenever I need help, your videos come through clutch!! Thank you so much for what you do!!
reply
Maurissio
I Have a 3080 with fan settings at 100% in MSI Afterburner, but T-Rex miner keeps showing only 80% of fan speed, what could be the problm
reply
I Have a 3080 with fan settings at 100% in MSI Afterburner, but T-Rex miner keeps showing only 80% of fan speed, what could be the problm
reply
Cole
I've watched a bunch of overclocking for mining videos, but this is the first time someone actually EXPLAINED it. lol. Thank you.
reply
I've watched a bunch of overclocking for mining videos, but this is the first time someone actually EXPLAINED it. lol. Thank you.
reply
joan
Hi Seb.
Did you try to undervolt your GPUs? It seems to lower VRAM temperature while keeping same MH.
But I didnt try yet.
reply
Hi Seb.
Did you try to undervolt your GPUs? It seems to lower VRAM temperature while keeping same MH.
But I didnt try yet.
reply
Add a review, comment
Other channel videos
