The computing power of my graphics card has been reduced by 5%. Who touched my graphics card's computing power?

Recently, Kuaikuai received a question from a reader about the loss of computing power of graphics card mining: My graphics card mining machine is connected to the Spark Mining Pool. During the mining process, I found that the actual computing power is always a few percentage points lower than the local computing power, and the submitted shares always have delayed shares and invalid shares. What is the reason for this? Has someone touched my computing power?

In order to obtain professional answers, Zhikuang University interviewed Qiu Xiaodong, the business director of Xinghuo Mining Pool, and invited him to provide in-depth answers to questions related to the computing power of graphics card mining.

The following is a summary of the interview.

01 Computing power and computing power difference

The Spark Pool page displays 4 values ​​related to computing power, as shown above.

Real-time computing power refers to the web computing power of the mining machine in the last 10 minutes, that is, the computing power that can generate actual mining income, representing the latest operating status of the mining machine . Real-time computing power is affected by various incidental situations and fluctuates greatly.

The average daily computing power refers to the average of the mining machine's web computing power over 24 hours, and is the basis for measuring the efficiency of the mining machine and whether its operation is stable .

Local computing power refers to the computing power of the mining machine locally, which is the computing power estimated by the mining software , that is, the theoretical maximum computing power without considering network delays and overclocking. Local computing power is the foundation, and only when the local computing power is high can the actual computing power be high.

The daily average local computing power refers to the average of the local computing power over 24 hours.

The ratio in the green label in the upper right corner of the above figure is the computing power difference, which refers to the difference between the daily average computing power and the daily average local computing power and the daily average local computing power ratio . The corresponding value in the figure is (19.41-19.73)/19.73=-1.62%. It is recommended to observe this value after running for a full 24 hours. The smaller this value is, the better the mining machine is optimized. It is normal within -5% .

02 Reasons for the difference between daily average computing power and daily average local computing power

Why is the daily average computing power lower than the daily average local computing power? This is because the daily average local computing power is reduced to the daily average computing power that can generate actual mining income. There are mainly the following computing power losses:

1. Mining software charges: generally 1%~2%

2. Network transmission loss: generally 2%~3%

3. Calculation task switching loss: 0.5%~1%

4. Severe overclocking loss: generally more than 1%

5. OS pumping: Not all OS charge by pumping computing power, and it needs to be analyzed based on the actual situation.

Mining software refers to the mining kernel (which will be described in detail below). The local computing power and the daily average local computing power are the computing power read by the mining kernel and reported to the Spark Mining Pool, which is displayed intact by the Spark Mining Pool . Miners use mining software to pay for the use of a certain percentage of computing power. This part of the computing power taken away by the mining software will be reflected in the difference between the daily average computing power and the daily average local computing power.

The daily average computing power has an additional network transmission link compared to the daily average local computing power, and the network transmission loss will also be reflected in the difference between the two .

Ethereum produces a block every 13 seconds on average. Every time a block is produced, the mining machine's computing task must be switched. During each switching of computing tasks, there will be a corresponding loss of computing power.

The result of severe overclocking is that a single card frequently and repeatedly submits shares to the mining pool , which will cause the server to not recognize it. The actual effect is the same as if the card is dropped, so it will seriously affect the web page computing power.

The OS cluster management system not only provides miners with a friendly interactive interface for observing and managing mining machines, but also optimizes the adaptation of accessories such as graphics cards, motherboards, and power supplies, which also helps to improve computing power. OS service providers will also charge corresponding service fees. The OS charging strategies are quite diverse. Some OS service providers charge for computing power, and some OS service providers may require miners to recharge USDT or ETH to collect copyright fees . The OS strategy for charging computing power can be divided into invisible charging of customers like mining software or cooperating with mining software service providers to share profits according to a certain proportion, etc.

In short, the local computing power minus the mining kernel's pumping, network transmission loss, computing task switching loss, severe overclocking loss, OS pumping, etc., can produce the web computing power that can generate actual mining revenue .

What methods can be used to optimize local computing power and web computing power? Let us introduce the factors that affect the local computing power and web computing power of mining machines.

03Main factors affecting the local computing power of graphics card mining machines

The factors that affect local computing power are mainly the following:

1. Graphics card performance

2. Bios parameters

3. Mining Core

4. Mining OS

5. Video memory quality

6. Graphics card life

7. Ambient temperature

1. Graphics card performance

Graphics card performance is the most important factor affecting local computing power. The better the performance of the graphics card, the higher the local computing power. For example, the performance of the Sapphire 5700XT graphics card is better than that of the 5600XT, so the local computing power of the 5700XT graphics card is higher. This is easy to understand.

2. BIOS parameters

BIOS is an important factor affecting the local computing power of the graphics card. BIOS is the abbreviation of Basic Input Output System. Graphics card BIOS, also known as VGA BIOS, is mainly used to store the control program between the display chip and the driver. The graphics card BIOS is solidified in a dedicated memory on the display card. The graphics card BIOS is mainly used for the control and management of the normal operation of various components on the graphics card, so the technical quality (rationality and function) of the BIOS program will inevitably affect the final product technical characteristics of the graphics card, and the effect of BIOS on the graphics card computing power is very significant .

It should be noted that the optimization space of BIOS is significantly different for A cards (AMD graphics cards) and N cards (NVIDIA graphics cards) . The design of A cards is open source, which means that all parameters of A cards can be debugged by DIY players themselves; the BIOS of N cards is locked, and N cards only have so-called soft overclocking, such as P104 cards. Users can only simply adjust the speed and frequency of external fans, do simple overclocking and downclocking, and use temperature to affect calculations. In comparison, A cards have too many adjustable parameters, leaving more room for miners to play.

3. Mining Core

Another factor that affects local computing power is the mining kernel (miner). The mining kernel is the software that guides the mining machine's calculations. Its function is to obtain computing tasks and assign them to the mining machine for scheduling and calculation.

The impact of miner on local computing power is reflected in its optimization of mining algorithms.

Miners can only mine with a mining kernel, and a good mining kernel can greatly improve the efficiency and stability of mining . As long as a competitive miner is posted on GitHub, someone will spontaneously make the miner compatible with various OS for miners to use.

A few months ago, the main miner for mining Ethereum was Claymore, which had the advantages of good stability, reliability, and low pumping, and its market share once exceeded 90%. After that, the development team stopped maintaining and updating Claymore. As Ethereum's DAG file continued to grow and the block height increased, Claymore could no longer mine. Currently, the competition among miners in the market is fierce, and Phoenix is ​​the leader in market share.

The mining kernel is the mining software that guides the mining machine's operations. A good mining kernel can greatly improve the mining machine's computing efficiency and stability, thereby improving the local computing power .

4. OS

OS is a cluster management system written based on Linux. The relationship between OS and mining machines is similar to the relationship between computers and operating systems such as Windows. A few years ago, there were no various OSs. When people configured mining, they directly used the Windows system and downloaded a Claymore or other mining kernel to interact with the Ethereum network.

Miner is a software that mainly interacts with the mining machine for calculations, while OS is a cluster management scheduling and monitoring system that can integrate various drivers, various mining algorithms and the miners corresponding to the mining algorithms.

In other words, the hash operation-related interactions of the mining machine are completed through the miner, and the OS system can install various miner software and specify a specific miner to run .

Graphics card mining is not just about the graphics card working, it also needs to work with the motherboard and power supply. The power supply, the motherboard transmits data, and the graphics card performs calculations. Only when all components work well together can there be higher computing power. The degree of adaptation of the OS cluster management system to accessories such as graphics cards, motherboards and power supplies has a great impact on computing power .

5. Graphics card quality

There are many main factors that affect the performance of graphics cards, the most important of which are the graphics card brand and the quality of video memory particles.

The general consensus in the industry is that graphics card brands are divided into three categories. First-tier brands include: Sapphire, ASUS, MSI, etc.; second-tier brands include: GALAXY, Colorful, Gainward, etc.; third-tier brands are ODM white-label. Generally speaking, the better the brand, the better the graphics card quality.

The memory chips also depend on the brand. Samsung and Micron are the most recognized, followed by Hynix, Elpida, etc. Of course, the specifics also depend on the batch and batch number.

Why does the video memory particle have a huge impact on Ethereum mining? This is because the GPU is far from running at full capacity during Ethereum mining, and only about 20% of its computing power is used, and 80% of the computing potential is not used. The 20% of GPU computing power that is used is mainly used to interact with the video memory in the DAG (directed acyclic graph). The interaction efficiency mainly depends on the video memory, so the quality of the video memory particles has a significant impact on the graphics card computing power .

6. Graphics card life

The life of the graphics card is also a factor that affects the local computing power. Taking NVIDIA's P104 as an example, as its service life increases, the speed will drop. This is because the best application scenarios for different NVIDIA cards are also different. Some cards are mainly designed for games, and some cards are designed for AI computing. Each batch of cards has good performance in the field they are good at, which is one of the reasons why NVIDIA has a large market share. The P104 card used to mine Ethereum will show a disadvantage in the later stage. Although the P104 has 6GB of video memory, which is much larger than the current 4G DAG file of Ethereum, because the P104 is too restricted, the larger the DAG file, the longer the P104 is used, the more serious its speed drop. This is also a factor that affects the local computing power. This point has actually been mentioned in the BIOS section above. The essence is that the adjustable space of the N card is too restricted.

The service life also has an impact on the A card. The impact on the A card is mainly reflected in the fact that after long-term use, the core voltage will shift, affecting the stability of the graphics card and thus affecting the local computing power .

This is also the reason why many old miners are willing to buy 5600XT and 5700XT. They know that these cards will be the main force in Ethereum mining in the next cycle.

7. Ambient temperature

Ambient temperature is also an important factor affecting the local computing power of the graphics card. The higher the ambient temperature, the lower the graphics card computing power will be, or even shut down.

The main factors that affect local computing power are the above, but local computing power is not the computing power that generates income. The computing power that generates actual income is the web computing power. The conversion of local computing power into web computing power that generates mining income is mainly affected by the network and network equipment and facilities . The factors that affect web computing power are introduced below.

04Main factors affecting web page computing power

Web page computing power is the computing power that can really generate actual benefits. The main factors affecting web page computing power include local intranet, mining farm network, mining pool server, mining kernel, etc. Let's introduce them one by one.

1. Mining farm local intranet

The local intranet of a mining farm is generally composed of the following parts: a network switch at the front end, which is connected to a router, and the router is connected to each mining machine. The higher the bandwidth of the network switch, the better, but cost factors must also be considered.

Different OS have different requirements for switch bandwidth. This mainly depends on whether the OS is disk-based or diskless, and whether the OS is flashed in the switch or in each mining machine.

There are various OSs on the market, such as Flint OS, Miner OS, BMiner OS, and Kuangbao OS. Some OSs are flashed into the switch, and all the mining machines connected to the switch are controlled through the switch. This OS has high requirements for the bandwidth of the switch , because each step of the switch's instructions and reading needs to be supported by the network.

Some OS are flashed to the hard disk of each mining machine, such as Flint OS of Spark Mining Pool. This is just like our own home computer, the system is in the local computer, this kind of OS has less stringent requirements on switches and routers, and gigabit bandwidth is enough.

The switches and routers in the local intranet of the mining farm are different from the ordinary brand switches and routers we use at home. When the switches and routers used at home are transmitting data, even if the CPU of the router automatically restarts, it will not have much impact on the home Internet experience. However, if the switches and routers used for mining have delays, automatic restarts, etc., then this impact will be reflected in the web page computing power and the final revenue .

The interaction between the mining pool and the mining machine is essentially that after the mining pool gets the Ethereum network block generation task, it assigns the task to all the mining machines connected to the mining pool through the mining pool server. The mining machine quickly completes the computing task assigned by the mining pool and then quickly submits it to the mining pool. The routers and switches in the mining farm are transmitting a large number of small files 24 hours a day, so the uplink bandwidth requirements of the routers and switches are very high. If the load capacity of the switches and routers is not good, file packet loss is likely to occur .

File packet loss means that even if the mining machine calculates the result that meets the requirements of the mining pool, due to the limited carrying capacity of the switch or router, the result cannot be submitted to the mining pool in time or even lost. For example, if the timely submission rate is only 97%, then the remaining 3% is the delayed share or even the invalid share, and the corresponding reward for this part of the share will be greatly reduced or even non-existent .

The above process explains how the mining farm's intranet bandwidth and its supporting facilities affect the mining machine's web computing power.

2. Mining Network

The mining farm network refers to the network of telecommunications, mobile or other operators that the mining farm is connected to. In some cities, the fastest and most stable network may be provided by telecommunications, while in other cities, the best network may be provided by other network operators. Network operators are also a key factor that mining farms need to consider.

Another factor that affects the mining farm network is the public IP. The public IP means that the mining farm network will not have too many jumps. The mining farm owner goes to the operator in the city to ask for a public IP for mining. The operator anchors the mining farm's IP, and the number of jumps in the middle of the transmission to the mining pool server will be reduced. Reducing the number of jumps can effectively improve the efficiency of mining machine submission shares and reduce delayed shares and invalid shares .

In addition , mining farms generally need to have backup broadband in case of emergency; mining farm networks are mostly used to submit shares to mining pools. Compared with home networks, they have higher upstream bandwidth and lower download bandwidth requirements .

3. Mining pool server

The mining pool server is also one of the important factors affecting the web page computing power. Now many mining pool servers are cloud servers. Cloud servers have dedicated hosting rooms and supporting high-defense, which are simple and convenient to operate. More importantly, excellent cloud server platforms provide high-speed channels covering the entire country, which is a service that self-built physical servers are difficult to provide. High-speed channels can reduce latency and help improve web page computing power .

Of course, the cost of high-speed channels is relatively high. In this regard, large mining pools can spread the cost through scale advantages, which is difficult for small mining pools to do. Therefore , the high-speed channels of large mining pools can help users improve web computing power .

4. Mining Core

The impact of the mining kernel on web computing power is reflected in the mining kernel's pumping.

The miner developer will charge a commission to the miners that use the miner. When your graphics card is mining, the miner may cut the computing power for 36 seconds (1%) every hour, and the output of this part of the computing power will also be sent to the developer's address. Different miners charge different commissions, ranging from 2%, 1.5%, 1%, to 0.5%.

The daily average local computing power given by Spark Pool is the computing power data given by the mining kernel, which is displayed intact. The part of the miner's deduction will be reflected in the web page computing power . If there is no miner deduction, and ignoring other various web page computing power losses, the original daily average web page computing power is 100MH/s. After the miner deducts 1%, the daily average web page computing power becomes 99MH/s.

When using a miner normally, the pumping part is basically not optimizable, unless you switch to a miner with lower pumping.

The above are the main factors that affect the daily average local computing power and the daily average web page computing power. These factors are also the main directions for optimizing computing power and computing power difference.

I didn't expect such a small question to lead to such a professional and comprehensive answer, which shows the professionalism of Spark Mining Pool in the field of graphics card mining. Thanks again to Qiu Xiaodong of Spark Mining Pool for his support.

Risk warning : The content of this article is only the author’s personal opinion, does not represent the views or position of Zhikuang University, and does not constitute any investment opinion or recommendation.