News GPU Mining No Longer Profitable After Ethereum Merge

These are all negative profit, most places are well above 10 cents per kwh due to the energy crisis caused by the Ukraine war. Europe is like 30-60 cents per kwh, I'm paying 14 cents. If you redo the math, all these cards are negative profit. You're literally spending more on power than you get back in coin. So you might as well just buy the coin with cash instead of trying to mine it.
 
Sadly what will be seen now, will be a bunch of crypto bros who ran their cards into the ground, now acting like used car salespeople who try to sell former rental car (fleet cars). It is common knowledge that those cars are abused and rarely get serviced, since those vehicles can technically function without routine maintenance, they just won't last very long, and routine maintenance means downtime which cuts into profits. If you want to test yourself, next time you rent a car, check the oil, odds are that it will be missing 1-1./5 quarts, since newer engines use lower tension piston rings to improve efficiency, and thus burn oil, but when the oil is only change once probably every 18 months in a rental fleet, they are almost always chronically low on oil. Then when it is time to choice the oil, they add some engine cleaning solution to the oil, run it for a while, then flush out the oil before adding new oil, and then dumping the car on the used market.
With all of that in mind, when it is time to try and sell them on the used market, they will make it seem like the cars were meticulously maintained, and driven only in the gentlest way possible.

Sadly a common trend behind the scenes at mining farms, (small and large scale, was to maximize hash rate vs TBP). This often means keeping the cards near their limits by using fan curves that keep the cards near their thermal limits since the fans on the card can easily pull 20-25 watts at max speed. Basically every watt that is not directly generating Eth, is seen as a waste that must be kept as low as possible.

Due to the compute arms race, the goal was not slow and steady, instead it was a rush for compute while trying to keep TBP low, and the easiest method, was keeping fan speed low.
Thus you will end up with cards after a few years where the VRAM is unstable because it spent the last year+ running at 110C, and the GPU within 5C of its thermal limit (before the card starts to cut back on its primary frequency table), as whether they mine or not, the difficulty will increase, thus there is a desire to maximize hash rate, but without entering the realm of diminishing returns.
People who end up buying those cards will likely end up with artifacting on their screen after a year or so of use.

A common misconception is people taking the false narrative from mining scalpers that they undervolt and underclock the cards and treat them delicately (the narrative is reinforced due to how widespread mining became and how many people on various sites and forums will be looking to offload those old mining cards, and they are not going to argue against their own interest. It is the GPU equivalent of a company that says a lemon is "certified preowned". One can even consider them to be along the mental lines of the NFT bros in how they will defend that crap, even though they know the issues with NFTs.
Mining cards were not babied; they were run into the ground and kept just barely operational, they were then cleaned at the last minute and dumped on ebay.
 
A common misconception is people taking the false narrative from mining scalpers that they undervolt and underclock the cards and treat them delicately (the narrative is reinforced due to how widespread mining became and how many people on various sites and forums will be looking to offload those old mining cards, and they are not going to argue against their own interest. It is the GPU equivalent of a company that says a lemon is "certified preowned". One can even consider them to be along the mental lines of the NFT bros in how they will defend that crap, even though they know the issues with NFTs.
Mining cards were not babied; they were run into the ground and kept just barely operational, they were then cleaned at the last minute and dumped on ebay.
They are under volted and under clocked for the same reason you claim they lower fan speed: to minimize power draw and maximize profit. And doing so usually lets the fans run slower as well.

The VRAM will usually be OC'd and ran hard though. I have no idea how likely VRAM failures are.
 
Last edited:
They are under volted and under clocked for the same reason you claim they lower fan speed: to minimize power draw and maximize profit. And doing so usually lets the fans run slower as well.

The VRAM will usually be OC'd and ran hard though. I have no idea how likely VRAM failures are.

While some will try to undervolt, it is not done to the detriment of the clock speed. Often the point at which you will get the highest hash rate per watt pulled by the GPU and VRAM, will be far away from the peak performance of the card, such a move would require giving up most of the performance of the card. When someone does a large undertaking of setting up a mining farm, they are not only looking at instantaneous profit per watt, they are also extrapolating future trends in mining effectiveness as the difficulty increases, as well as changes that come from other miners increasing their overall compute. That arms race, forced miners who are looking to maximize long term profits, to balance peak hash rate and TBP.

The current solution is effectively keeping TBP as close to the GPU and VRAM power draw as possible. This often leads to a small VRAM overclock, GPU undervolting if possible without losing clock speed, if not they will stay on stock, and then having a fan curve that keeps the GPU on the edge of throttling, and if needed, taking steps to level thermals, e.g., if a very low fan speed leads to ineffective cooling for the VRAM and some modules are starting to hit like 115C and corrupt data, they might zip tie a small heatsink with a blob of thermal paste to a specific part of the back plate, and then point a fan at it, if they can save 4 watts on the main fans by having a 1 watt fan targeted as a specific area of the back plate, thus allowing for more leveled thermals so that everything can be kept closer to their thermal limits.

The mindset of the GPU miners is totally different from those who enjoy PC building and gaming where we will put in extra effort to lower our temperatures by an extra 5-10 C even if that reduction is not directly giving extra performance, of if the extra cooling will allof for an ever so slight performance boost that is extremely deep into the realm of diminishing returns on terms of performance per watt for total system power draw.

The GPU miner mindset is one of a projected profitability life, and a goal to keep the card alive (barely) to last the length of the mining lifecycle, and if the card is still alive afterwards, they will clean it up and dump it on ebay. Just like the rental car fleets where they buy a car, and never change the oil for like 2 years, then doing an engine flush and adding some oil before dumping it on a used car lot where the salesperson will claim that the car taken care of so well that all life in the rest of the galaxy retroactively disappeared because they were so jealous of how well maintained the car was.
 
I'd like to be proven wrong in the future, but I suspect mining companies, even AIB partners, but particularly Nvidia, will definitely find a way to prop up a crypto coin or two to continue.

It'll be a planned, engineered and carefully sustained half-craze, looking oh-so-spontaneous and unexpected, with a planned trickle of new GPUs by Nvidia, to keep supply low and prices high, exactly like the best part - for them- from past period.

Engineered, predictable and ultra profitable, with no negatives from this last period, so no scalpers, no real shortage of components or anything similar. Nvidia's revenues are so dependent on mining that I really don't expect them to sit and wait for a 'maybe'.

They'll find a way to prop up a coin or several, launch a hype campaign, start another pyramid and bubble etc, with untraceable transactions, with help from mining 'investors', and mining companies, to keep it going. We're totally naive thinking it will be different this time.
 
I'd like to be proven wrong in the future, but I suspect mining companies, even AIB partners, but particularly Nvidia, will definitely find a way to prop up a crypto coin or two to continue.

It'll be a planned, engineered and carefully sustained half-craze, looking oh-so-spontaneous and unexpected, with a planned trickle of new GPUs by Nvidia, to keep supply low and prices high, exactly like the best part - for them- from past period.

Engineered, predictable and ultra profitable, with no negatives from this last period, so no scalpers, no real shortage of components or anything similar. Nvidia's revenues are so dependent on mining that I really don't expect them to sit and wait for a 'maybe'.

They'll find a way to prop up a coin or several, launch a hype campaign, start another pyramid and bubble etc, with untraceable transactions, with help from mining 'investors', and mining companies, to keep it going. We're totally naive thinking it will be different this time.
Nope, this is a bad idea. Nvidia wants stable business, and so do its partners. The mining boom and bust cycle is terrible, as there's no predictability. This is probably the number one reason EVGA is exiting the graphics card business. Building BTC ASICs would be much better as a business plan than trying to create new coins that will be GPU mineable — from Nvidia's perspective. BTC has more or less reached equilibrium and while price fluctuates a lot, the general trend is simply higher hash rates. And BTC hasn't expressed any intention to fundamentally alter the underlying algorithm — although there have been several hard forks of BTC over time.

I don't think Nvidia's LHR initiative was fake. It was intentional, but it was also slapped on to existing designs. Nvidia didn't design Ampere to be good at mining; it was a fast gaming GPU that just happened to work really well for mining. RTX 40-series will be moving to more L2 cache without massively increasing memory bandwidth, which means it won't be substantially faster for mining than RTX 30-series. But ETH mining is dead now, and GPU mining profitability is negative on nearly every potential algorithm, coin, and GPU for the time being.

As I pointed out, the network hashrate for ETH was over 900 TH/s for most of the past nine months prior to The Merge. That's equivalent to roughly 10 million RTX 3080 GPUs, and of course there was a big mix of older and slower GPUs along with a few faster RTX 3090/3090 Ti cards. Conservatively, probably 20 million GPUs of varying speeds were mining Ethereum prior to the Merge. By contrast, something like BitcoinZ (Zhash algorithm) was sitting at ~70 KH/s prior to the merge. That's the equivalent of roughly 520 RTX 3080 cards, or maybe a mix of 1,000 different GPUs. Since The Merge, the BTCZ hashrate has shot up to ~240 KH/s, which is equal to 1,800 RTX 3080 cards.

Or maybe everyone with an AMD GPU should shift to Grin-CT32 (GRIN) for mining? That's technically profitable on the RX 6800/6800 XT if your power is cheap enough. Network hashrate is about 10K graphs/s, with a 6800 doing 0.9 G/s. So that's maybe 11,000 AMD Navi 21 GPUs. Octopus was another algorithm that some suggested might take the place of Ethash after ETH went PoS. Conflux (CFX) is the best coin on that algo right now, with a network hashrate of about 3.3 TH/s and RTX 3080 does 77 MH/s, so that's equal to perhaps 45,000 GPUs.

You see the problem? That's a several orders of magnitude difference in GPU quantity. 20 million GPUs compared to, at most, maybe 100,000 GPUs on the current "best" options.

Incidentally, Ethereum Classic (ETC) saw it's network hashrate go from around 50 TH/s to over 230 TH/s — all those ETH miners obviously thought/think Ethereum Classic is the next best option. But the price of ETC hasn't changed really, sitting around $35 per ETC. At a rate of 3.2 ETC per block and 15 seconds per block, that's 18,432 ETC per day, or $645,000. 230 TH/s is a lot, though, which means an RTX 3080 might get about $0.25 per day in ETC while consuming $0.55 in power at $0.10 per kWh. And if you're paying $0.20 per kWh, then ETC would drop you to losing almost $1 per day.

TL;DR: It will take a "miracle coin" to repeat what Ethereum has done with GPU mining. I'm hopeful it simply won't happen at all.
 
  • Like
Reactions: Krotow
While some will try to undervolt, it is not done to the detriment of the clock speed. Often the point at which you will get the highest hash rate per watt pulled by the GPU and VRAM, will be far away from the peak performance of the card, such a move would require giving up most of the performance of the card. When someone does a large undertaking of setting up a mining farm, they are not only looking at instantaneous profit per watt, they are also extrapolating future trends in mining effectiveness as the difficulty increases, as well as changes that come from other miners increasing their overall compute. That arms race, forced miners who are looking to maximize long term profits, to balance peak hash rate and TBP.
Anyone "properly" mining ethereum underclocked and undervolted their card for maximum profitability. Read any ethereum mining guide and they will tell you to underclock. I had my 3090 underclocked by 500MHz when mining. Any drop in hashrates was more than offset by the reduction in power usage.
 
While some will try to undervolt, it is not done to the detriment of the clock speed. Often the point at which you will get the highest hash rate per watt pulled by the GPU and VRAM, will be far away from the peak performance of the card, such a move would require giving up most of the performance of the card. When someone does a large undertaking of setting up a mining farm, they are not only looking at instantaneous profit per watt, they are also extrapolating future trends in mining effectiveness as the difficulty increases, as well as changes that come from other miners increasing their overall compute. That arms race, forced miners who are looking to maximize long term profits, to balance peak hash rate and TBP.

The current solution is effectively keeping TBP as close to the GPU and VRAM power draw as possible. This often leads to a small VRAM overclock, GPU undervolting if possible without losing clock speed, if not they will stay on stock, and then having a fan curve that keeps the GPU on the edge of throttling, and if needed, taking steps to level thermals, e.g., if a very low fan speed leads to ineffective cooling for the VRAM and some modules are starting to hit like 115C and corrupt data, they might zip tie a small heatsink with a blob of thermal paste to a specific part of the back plate, and then point a fan at it, if they can save 4 watts on the main fans by having a 1 watt fan targeted as a specific area of the back plate, thus allowing for more leveled thermals so that everything can be kept closer to their thermal limits.

The mindset of the GPU miners is totally different from those who enjoy PC building and gaming where we will put in extra effort to lower our temperatures by an extra 5-10 C even if that reduction is not directly giving extra performance, of if the extra cooling will allof for an ever so slight performance boost that is extremely deep into the realm of diminishing returns on terms of performance per watt for total system power draw.

The GPU miner mindset is one of a projected profitability life, and a goal to keep the card alive (barely) to last the length of the mining lifecycle, and if the card is still alive afterwards, they will clean it up and dump it on ebay. Just like the rental car fleets where they buy a car, and never change the oil for like 2 years, then doing an engine flush and adding some oil before dumping it on a used car lot where the salesperson will claim that the car taken care of so well that all life in the rest of the galaxy retroactively disappeared because they were so jealous of how well maintained the car was.
Most GPU mining is (was) ethereum mining, which is heavily memory bandwidth bound. You can down clock the GPU (to a point) with essentially no loss on hash rate. Every thing I've read (plus my own experience dabbling in mining back in ~2017-2018) suggests that down clocking the GPU is standard procedure. Or if you're not explicitly lowering the GPU frequency, you lower the voltage and/or power limits, which will automatically lower the operating frequency in modern cards.

TH has an article about tuning your cards for mining, which discusses dropping core clock/power limits:
https://www.tomshardware.com/how-to/optimize-your-gpu-for-ethereum-mining
 
  • Like
Reactions: JarredWaltonGPU
This is fine, I got my money back from the gpus 😁
The 3080 will replace my aging r380.
This is fine, I got my money back from the gpus 😁
The 3080 will replace my aging r380.

More than fine here too. one 3080 ti, one 6800xt, five 6800's, one 5700x all payed for by mining. Now...which one to game with...eenie meenie minee(no pun intended) mo. Maybe i'll switch out the 3080 and the 6800xt at my whim while reading gamer sob stories and sipping coffee. I see overstocked shelves at Microcenter now, so where are all these poor deprived gamers to buy up the cards? While the gamers whined i mined...lol
 
More than fine here too. one 3080 ti, one 6800xt, five 6800's, one 5700x all payed for by mining. Now...which one to game with...eenie meenie minee(no pun intended) mo. Maybe i'll switch out the 3080 and the 6800xt at my whim while reading gamer sob stories and sipping coffee. I see overstocked shelves at Microcenter now, so where are all these poor deprived gamers to buy up the cards? While the gamers whined i mined...lol
Way to live up to the cliché.