Mining means providing computing power for the cryptocurrency network protection in exchange for a potential reward for equipment operation. Sadly, many legislators wrongly believe that mining is directly connected to coin creation. It’s wrong to say that miners issue cryptocurrency or somehow participate in this process. For most modern currencies the daily outflow of up to 10% of miners will affect the speed of coin issuance by less than 1%.
If there are only ten miners out of a million left in the end, the coin creation will not stop and soon regain the same level with adjusted difficulty.
Miners provide protection to cryptocurrencies against the potential hold or cancelation of a transaction. More miners mean better protection and lower risk of an attack. They would make an attack almost impossible due to the lack of necessary mining equipment in the world.
Any PoW cryptocurrency network (Proof of Work) pays miners for this work with their coins by issuing new coins and using transaction fees.
Let’s draw an analogy with the modern banking system. Cryptocurrencies are the Central Bank (issues money), mining pools are commercial banks (prepare transactions, distribute rewards, and compete with each other), and miners are clerks (does the essential work). Banks can’t operate without clerks, but it doesn’t mean that clerks issue money.
Contents
How to Mine Cryptocurrencies in 2021
Who can we consider a professional miner? How much experience must they have? We don’t have an exact answer, but a professional miner is someone who knows how to adjust under constantly changing circumstances. The crypto market is highly volatile. A certain coin can be 20–30% less profitable than your optimal coin today, but bring twice as much profit tomorrow.
Here’s an example. Cortex (CTXC) was bringing $4.17 per GPU (1080Ti) on August 13–15, 2020, two weeks later – less than $1.8 a day, and on August 30–31 it rose to $3.31 a day per GPU.
Mining is a source of passive income for most miners, so they usually don’t switch from one coin to another on a daily basis. They prefer to stick to major, expensive coins with a stable average profit.
However, every miner must know the capabilities and limits of their graphics cards.
Sure, there are profitability calculators, so why would you need to know that? Even 2CryptoCalc, a powerful cryptocurrency profitability calculator, can’t tell you how an increase in your utility rates will affect mining profitability, or whether the profitability of your coin will decrease five or more times tomorrow. It also can’t tell you whether the room is ventilated well enough for mining, when a multi-rate meter switches rates, and which equipment noise level is acceptable during the day or at night.
ASICs and GPUs for Mining
Initially, CPUs were considered a universal mining tool that everyone could use on their computer. Soon after, it became clear that GPUs are much more powerful and efficient, while single-purpose devices (rare FPGAs and widespread ASICs) are even more powerful.
ASIC is a single-purpose mining device designed for a specific algorithm. ASICs are produced by a few companies with highly opportunistic pricing policies. They are also notorious for mining on their devices before shipping them to clients.
ASIC mining is straightforward, but it doesn’t provide a stable profit. Competition is limited only by operation capacity and manufacturers’ greed.
Cheap and powerful ASICs replaced GPUs on most old algorithms unwilling to protect themselves from such devices.
The biggest ASIC issue is its limited use and constantly decreasing cryptocurrency profit as manufacturers are constantly launching newer, much more energy-efficient models.
A Graphics card (GPU) is a widespread universal device that works with multiple new algorithms and has a good depreciation value (40–60%) on the second-hand market even after a few years of use. You can sell graphics cards to gamers after the warranty expires (or, even better, right before).
Most new cryptocurrencies are trying to be as decentralized as possible protecting themselves against ASICs. Plus, you can improve GPU energy efficiency multiple times even within the same GPU series. The pattern is different for every algorithm. So, apart from GPUs, miners’ assets include valuable knowledge they acquire in the process and relevant information they obtain about the crypto market.
Plus, GPUs with limited power bring several times more profit per kW than ASICs.
So for personal use, miners usually use GPUs of the latest or previous generation.
Graphics Cards for Mining
AMD and Nvidia are two major worldwide corporations that produce powerful gaming GPUs. Miners and gamers call them “red” and “green” respectively (by the color of their logo).
Mining utilizes computing rather than gaming GPU capabilities using special libraries (OpenCL for AMD and CUDA for Nvidia). Such use of graphics cards is considered non-standard. GPU load in mining is similar to its load when designing and calculating same-type texture maps in gaming. With more gaming power added due to growing gamer demands, GPUs acquire more computing power.
AMD Mining and Overclocking Basics
AMD GPU architecture was initially designed to have fewer logic blocks and more stream processors, so it worked better with the first mining algorithms successfully executing a lot of minute operations. It’s no surprise that before the launch of ASICs AMD cards accounted for 95% of mining. The rest were mining on processors.
Modern mining algorithms of new coins require good interaction between core and memory which is beautifully implemented by gaming GPUs, but it’s hard to implement for ASICs. Historically, Ethereum (ETH) has become the most successful ASIC-resistant project among new cryptocurrencies. AMD cards were designed better for ETH mining than Nvidia.
These two facts have defined mining for years to come. From 2015 through 2018, AMD GPUs had a major share in mining, and in most cases, they were mining Ethereum.
The basic setup of AMD cards for Ethereum mining is quite simple.
- First, we should identify the max memory clock at +25 MHz intervals. Find the clock value that makes it crash – reload – the GPU driver, and move 50 MHz down, to the stable area.
- Find the core voltage that will give you the desired GPU energy consumption. Lower voltage results in lower consumption, but also in a lower max hash rate. The relationship is not proportional. When you lower the consumption by 20–25% from the standard, the hash rate decreases only by 10–15%.
- and the max stable core clock (at +25 MHz intervals), move 25 MHz down from it.
Fine-tuning would give us an additional 5–6% of hash rate or 3–4% of energy efficiency, but we will not go into more detail in this article.
Nvidia Mining Basics
The basic Nvidia GPU architecture had a significantly lower number of stream processors combined in larger CUDA blocks which made them less efficient for mining until 2016.
The 9xx series with the Maxwell architecture could already mine ETH at an acceptable hash rate. However, it was the new 10xx series with the Pascal architecture in 2016 to bring to an end AMD’s dominance in mining.
Gamers have always appreciated Nvidia graphics cards for their energy efficiency and good cooling systems. They weren’t heating like AMD cards and weren’t dropping clocks in games due to overheating. The green brand has become associated with optimal energy consumption. With every new series, Nvidia was trying to get ahead of AMD on this parameter.
The 10xx series was no exception. In 2016–2018 it was still behind AMD in terms of Ethereum mining power, but it was leading in energy efficiency. Nvidia on the 16 nm process achieved better results than AMD on 14 nm (the 4xx and 5xx series).
Important: Nvidia GPUs show full consumption (not just core consumption), so you can’t compare them directly to AMD GPUs using the information from drivers.
You can compare them only by measuring real consumption when you plug the whole rig.
The 10xx series wasn’t designed for mining, so in 2020 miners with the 10xx cards experienced a drop in hash rate on Ethash (the current Ethereum mining algorithm). Miners had to increase energy consumption to maintain the hash rate which affected energy efficiency. AMD encountered that problem at the beginning of 2017. They solved it by producing special drivers. Then they integrated a special Compute mode in standard drivers. Nvidia solved this problem only for their new series (16xx and 20xx).
Despite the problem of the 10xx series with Ether mining, green cards still have a significant advantage. They are better suited for many other mining algorithms. Most cryptocurrencies that emerged after 2017 didn’t want to use existing algorithms. They wanted to create somewhat unique products.
Before we get to algorithms for green GPUs and Nvidia overclocking principles, we are going to address an important question: how to assemble a working computer for mining with several graphics cards, known as a “rig” among miners.
How to Build a Mining Rig and Choose an OS
We are not going to talk about hardware. All modern motherboards support mining with at least 4 GPUs (most support 8+ GPUs with cheap slot extenders).
We are mostly interested in operating systems. Most miners can be divided into two groups: Windows and Linux users. The latter mostly don’t use Linux in its original form. They use ready-made mining operating systems, like HiveOS, RaveOS, and mmpOS.
Rave OS
Rave OS is a system for miners that allows them to control mining rigs remotely. Rave OS promises stable mining, high hash rate, and minimal down-time that usually leads to losses.
You can manage three devices for free. If you have more, you pay $2.00 a month per each.
The main advantage of the system is its simplicity. You just have to record an image on the data storage device and connect it to your device. The system automatically detects the model of the equipment, adjusts settings accordingly, and allows you to start mining.
Mining on Windows
Mining on Windows has a few main disadvantages.
- You need to install a lot of necessary mining programs and drivers manually, disable auto-updates and antiviruses.
- You need to study stability and reboot of mining devices in and out.
- There is no built-in restart algorithm in case the system crashes because of power supply issues, freezing GPUs, etc.
You also need to install a remote control/desktop system, as you can’t be around your rig all the time. Plus, rigs usually don’t have additional monitors, mouses or keyboards, as it would add expenses on peripherals.
There is an issue with Nvidia graphics cards on Windows. The GPU memory at a standard clock commits a few errors that are not critical when rendering images in games, but they may be critical when making exact calculations. So for CUDA, Nvidia designed a special lower memory clock state (P2).
GPUs switch to this state automatically, but miners often try to maximize hash rate by overclocking memory significantly (to +700, +1000, or even more). We have a situation where P2 lowers the clock, while a mining device increases it at the same time. The mining device keeps working, but when you turn it off, P2 switches back, while manually added overclock remains. This leads to clocks that GPUs can’t manage. As a result, the driver crashes when you turn off or restart the mining program.
The best thing to do is to disable CUDA P2 state through the nvInspector tool (more specifically, through its nvProfileInspector subtool). Find CUDA Force P2 state parameter, disable it, and launch your miner.
Important note: this parameter is reinstated every time you install a new Nvidia GPU, so don’t forget to double-check.
If you use MSI Afterburner to monitor your GPUs (every miner should use it), you will notice an increase in operating memory clock when mining and a higher hash rate when mining RAM-dependent algorithms, like Cortex. Memory clock will not fluctuate, so you can overclock memory safely. You will get a real operating value in MHz, not just a plus sign in the settings.
Nvidia Mining in 2021
Nvidia graphics cards are more universal in terms of algorithms, but it doesn’t mean that they are always more profitable than AMD.
Mining algorithms can be “hot” (the GPU gets really hot) or “cold” (the GPU doesn’t heat up too much). Hot algorithms mostly use GPU core, so their performance depends on the core clock. The max core clock is achieved by the increasing voltage which leads to a proportional increase in GPU power consumption.
Cold algorithms mostly use memory, so they are highly limited by bandwidth capacity or memory access time. As a result, there is no point in overclocking the core much, as it doesn’t increase the hash rate. Maximum efficiency is easily reached on such algorithms. We need to find out which core clock maximizes hash rate, and then obtain that core clock with the lowest possible stable core voltage.
In summer 2020, green cards didn’t have a lot of algorithms. The leading algorithms (coins) by profit were: Ethash (ETH), MTP (XZC), and Cuckaroo30 (CTXC).
Let’s get into more detail.
Nvidia Mining and Overclocking on Ethash
Ethereum algorithm used to be one of the most efficient for the Nvidia 10xx, 16xx, and 20xx series. However, every 3–4 days its difficulty rises by a bit, which requires more and more GPU memory. At first, the requirement was 2GB, but at the end of 2020, GPUs were required to have at least 4GB of RAM to mine ETH.
It turned out that the 10xx GPU architecture prevents them from mining ETH effectively on the new epoch. In a few months, their mining hash rate dropped by 25%. You can remedy this by increasing core clock and voltage by a lot which will lead to a significant decrease in energy efficiency.
Conclusion: from 2020 onwards, ETH mining on the 10xx GPUs is not the most efficient alternative.
The newer 16xx and 20xx GPUs don’t have this problem, so they can mine this cold algorithm efficiently. And with enough memory (6GB+), you won’t have to worry about the DAG file increase for a few years.
The most efficient configurations are below. This is the GTX 1660.
And this is the RTX 2070s.
As you can see, you don’t gain any hash rate by increasing core clock and voltage. So the optimal energy efficiency for Ethereum mining for this GPU series is 220W–230W per 100 MH.
Setting the core to 650 mV on new cards leads to a significant decrease in max hash rate, but doesn’t increase energy efficiency, because GPU switches to the P3 state with a significant decrease in memory clock (much needed for this algorithm).
Nvidia is getting closer to AMD by the number of stream processors, so the new series (the 30xx and modifications) became even more competitive on this algorithm.
However, Ethereum is planning to abandon the PoW system, so its profitability in the next few years may decrease by a lot.
Nvidia Mining and Overclocking on MTP
The MTP algorithm appeared when Zcoin was switching (or running away) from the old algorithm due to the ASIC threat.
The algorithm was initially designed as ASIC-resistant. It requires at least 6GB of GPU memory. Luckily, this parameter doesn’t grow as it does with Ethereum.
The algorithm is quite hot. It depends highly on both the core clock and the memory clock. You won’t necessarily increase energy efficiency by simply lowering the voltage.
As you can see in the screenshot if you lower the 1660 core voltage to 650 mV, the hash rate drops significantly, but energy efficiency doesn’t grow. It’s important to search for and know your GPU’s optimal minimum, as well as an optimal maximum (without overheating). Overheating leads to a significant deterioration of GPU functions and can even cause it to break.
The 10xx series efficiency ranges from 15kH/W to 22kH/W. The 20xx series ranges from 19kH/W to 3kH/W.
The new series must have improved this value to some extent (by 20–30%). They haven’t raised it multiple times (like with Ether), so even in 2020 the 10xx series GPUs are working well on this algorithm.
You need to activate the pill to maximize the hash rate for the 1080 and 1080ti (3.4MH and 4.4MH respectively).
Nvidia Mining and Overclocking on Cuckaroo30
The algorithm for Cortex is hybrid. It uses both core and memory with the latter prevailing. It can be considered a rather cold algorithm. It works well for mining on GPUs with at least 10GB of RAM. For example on the 1080ti it gives out as much as MTP does at full capacity or Ethash at standard capacity. However, the real energy consumption of plugged graphics cards is 40% lower than that of MTP and 25% lower than that of Ethereum.
It’s worth pointing out that Cuckaroo30 is an algorithm with a changing load. Sadly, systems like HiveOS only show the current load changing within a 30% range. In the mining program, say Gminer, you can see the real load, but most users interact with the web interface that doesn’t provide this information.
For example, if such users set the power limit at 240W in RaveOS for the 2070s GPUs, they wouldn’t understand that on this algorithm a GPU doesn’t consume more than 200W, as the short active core load stage is followed by the long memory load stage.
You can easily notice it in Windows. When you set the max PL at 111% (245W), the real-time consumption monitor will only show a few single values higher than 220W following one another. The rest of the time consumption is much lower, with a minimum of 158W and a mean value of 197W. Note that with Platinum PSUs, estimated power values correspond to the real consumption.
Windows 7 and Linux systems (RaveOS, MMP, etc.) allow you to use Gminer to mine with 8GB GPUs.
It’s not efficient for the 10xx series: the hash rate of 1070, 1070ti, and 1080 is 1.55, 1.5, and 1.65 respectively. But it works well with the 2070 and 2070s, where the hash rate is directly dependent on the Power Limit value. If you set PL at 240W, the 2070s with 8GB consumes around 197W and gives out a decent hash rate (comparable to the 1080ti with 11GB).
Mining Algorithms for Low-End Nvidia GPU’s
Sadly, GPUs with low memory, old GPUs, and even some new GPUs can’t mine the algorithms mentioned above, and if they can, they are less efficient. There are a lot of profitable algorithms for such Nvidia cards.
The two most profitable in the second half of 2020 are BeamHash-III (Beam) and Kawpow (Ravencoin).
Nvidia Mining and Overclocking on BeamHash-III
The third version of the Beam algorithm is an ASIC-resistant algorithm for GPU mining.
The initial requirement was 3GB+ GPUs, but soon after LolMiner activated support for 3GB GPUs (i.e., for memory that doesn’t display images in Windows).
When you overclock the core of the 10xx series GPUs, the algorithm gets limited by their memory pretty fast, at the average power consumption level (around 70%). Further core overclocking doesn’t lead to an increase in hash rate. So this algorithm can’t be considered hot, but it brings a decent average annual profit on 3GB/4GB GPUs.
The algorithm is not featured among the most profitable for 6GB+ GPUs , as there is a large number of old GPUs on the algorithm.
Nvidia Mining on Kawpow
Kawpow (with RVN as a major coin) was quite promising at the beginning, but after three months it wasn’t featured among the most profitable algorithms anymore.
That’s a short success story, but Kawpow has some key advantages. The algorithm works even with 3GB GPUs, adjusts well (many GPUs that are problematic on other algorithms work well on kawpow), the hash rate correlates to GPU PL, and RVN is featured on many major exchanges.
If you look at all the factors, the algorithm comes second best in terms of prominence and profitability after Ethash.
Overclocking is similar to Beam, so we won’t get into more detail. It was Beam to take on some part of 3GB GPUs, which allowed kawpow users to feel more optimistic in 2020.
Plus, after the ETC upgrade and the outflow of more 3GB GPUs there, Kawpow will be featured among the most profitable algorithms more often, even for powerful GPUs.
Nvidia Mining and Overclocking on Cuckarooz29 and Cuckatoo32
Both algorithms (C29 and C32) mine GRIN cryptocurrency that was initially said to be GPU-focused and ASIC-resistant. So it makes slight changes to algorithms from time to time to make it too costly to build new ASICs for the network.
Grin mining on C29 was featured among the most profitable algorithms very rarely. This algorithm was tied to a single coin, and Grin’s popularity went a bit down. Grin v5.0.0 hard fork has happened at block 1 048 320 (January, 15) and the C29 algorithm was terminated. At the moment C32 is the only active algorithm for GRIN mining. Read more
Unlike C29, C32 requires at least 8GB of GPU memory for mining, and top GPUs in the series give out a higher hash rate, just like with the Cortex algorithm. This algorithm is often featured among the most profitable for the 1080ti, 2080ti, 3060ti, 3070, 3080, and 3090.
The algorithm doesn’t cause issues and works well even on weak GPUs.
Nvidia Mining and Overclocking on Equihash-xxx
Anonymous ZEC coin on equihash was launched in 2016 challenging the ETH dominance in mining. Initially, coin mining was only available on CPU, but after a month they launched a GPU mining device that was giving much better results on green cards than on red cards.
Equihash is highly dependent on the core overclocking and doesn’t have large-size temporary files to place in GPU memory, so it was an easy target for ASICs. Despite the network centralization threat, they didn’t change the algorithm, which made people talk about their affiliation to Bitmain, the biggest ASIC manufacturer.
Unlike the main coin, its various modifications felt threatened by the growing hash rate that might have been concentrated in someone’s hands. So they decided to change the algorithm. That’s how multiple Equihash-xxx variations appeared. xxx is a numerical ID code for example Equihash 125,4 (used by ZelCash) or Equihash 144,5 (used by Bitcoin Gold).
At the same time, the main coin exchange rate fell significantly, as well as the crypto market as a whole. That turned the production of new ASICs for each modification into a losing game.
However, multiple coins created on these algorithms after 2016 can’t compete with top coins in terms of profitability.
If you decide to start mining these coins, you will have to monitor profitability and switch between algorithms/coins from time to time. With that being said, you could still make a nice profit (especially with old GPUs).
These algorithms mostly need core overclocking.
Energy Efficiency in Mining. Operating modes
Any GPU can operate in power-saving mode, but overclocking allows you to maximize the hash rate. At the same time, energy efficiency suffers.
Not all rigs work well with any algorithm, so every expert user must test their GPU operation at maximum and minimum capacity and choose the optimal mode, considering operating temperature and electricity costs.
You can save the results as links to the 2CryptoCalc mining profitability calculator that allows you in just a few clicks to evaluate whether your algorithm is better or worse than alternatives in terms of the current mining profitability specifically for your rigs.
Let’s take as an example a small but problematic rig that can mine well only four algorithms on Linux.
Below you can see the results for the middle of January 2021.
You can see that it made sense to overclock the rig in January only at the electricity cost of no more than $0.09 per kWh.
Hot/Cold mode electricity consumption difference is 190W.
That means the Hot mode requires 0.190 x 24 = 4.56kWh more every day.
The hot mode rig operation daily reward is $9.50 while the cold mode reward is $9.09. The hot mode gives you $0.41 more.
0.41 / 4.56 = $0.09 per kWh
Also, remember that the night power rate is usually lower so it would have made sense to operate at full capacity. Electricity costs and current profit are crucial when you are choosing an algorithm and rig operating mode.
Max Energy Efficiency for New Mining Rigs
When you build a new rig, you have to remember that GPU max energy efficiency within one series is almost the same. Usually, mid-range GPUs ($300–$400) have the best max power-to-price ratio.
But it’s tedious to build and monitor a large number of weak rigs, so miners usually choose GPUs at $500–$700.
Sooner or later, all miners end up estimating the max hash rate per 1 kW of a rig power due to either electricity cost or the “free” power limits.
Here are max hash rates adjusted for the dev fee per 1 kW of utility power with Platinum PSUs for the 10xx, 20xx, and 30xx series. The 30xx series stats are up-to-date as of December 25, 2020, so make sure to follow the updates.
As you can see the 20xx series has a marginal benefit per 1 kW 1.5x higher than the previous one. It’s mainly due to the enhanced memory access algorithm. The effect on the core algorithms is not so impressive (33%).
GPU Series Comparison. AMD 5xxx–6xxx vs Nvidia 20xx–30xx
Competition between AMD and Nvidia leads to the constant series upgrade between the two in an attempt to beat the competitor.
In 2017 AMD released the popular 4xx-5xx series. As a response, Nvidia released the 10xx series increasing Nvidia’s share on the market for several years in a row.
Only in 2020 AMD managed to launch the 57xx series in response not only to the 10xx, but also to the less popular 20xx (Nvidia had to upgrade it quickly to the Super subseries).
Nvidia 30xx and AMD 60xx surprised us with the increased number of stream processors. However, the power of core algorithms didn’t increase multi-fold for the first 30xx cards. The results of memory-focused algorithms, ETH in particular, are fantastic. The architecture was modified to a great extent, close to the modern technology limits. It means that multi-fold increase in power is impossible without a substantial rise in GPU prices.
We hope that the future series will feature new ways of interaction with GPU memory to use new and powerful cores to their fullest potential (like the 3090 and its modifications).
Conclusion
Despite the late start in the mining industry, Nvidia has obtained a significant share of the market. As of September 2020, the open data shows that green cards account for 39% of all video cards used for mining as opposed to 25% in 2018. 90% of those cards are from the 10xx series. The 20xx series is only entering the market, while miners may already shift en masse to the next 30xx series very soon.
According to 2Cryptocalc, the versatile choice of algorithms closes the gap between the top and alternative algorithms for Nvidia GPUs. Plus, there are many efficient algorithms to choose from.
We hope that this article will help beginning miners to maximize their profit without damaging or overheating their GPUs.
PS
At the beginning of 2021, we will see a lot of new GPUs. It means that the balance of power may change as new GPUs, mining programs, and algorithms are entering the market. So we will try to keep this article up to date as we get the new data.
Prepared by Denis Trushin, Head of the Advanced Technology Department at ADM Techno. Denis is an active member of the 2Miners mining pool community.
Remember to follow us on Twitter to get all the news as soon as possible.