Bytom Blockchain Protocol Works to Make Obsolete Mining Machines Useful Again With AI ASIC Chips
In 2009, with the inception of the Genesis Block, bitcoin adopted the proof of work system to address the byzantine general problems, which is like sacrifice efficiency for security.
The concept of POW was first presented by Cynthia Dwork and Moni Naor in a 1993 journal article. In the following years, it was accepted by more people as a system to prevent various forms of attacks of a network, such as denial of service or spam. A key feature of the system is that it forces someone connected to the network to perform a bit of computer work before they’re allowed to participate, which makes it expensive to attack the network. The work required attackers to perform can be divided into three groups in terms of computer resources.
- CPU resources. A good example is Hashcash that was created to limit the amount of email spam and DoS attacks on the Internet.
- Memory resources. For example, in order to avoid the possible 51% attack caused by the consensus mechanism of bitcoin, Ethereum adopted a kind of POW algorithms that requires more computer memory.
- Network resources. An attacker must collect some tokens from remote servers before he could perform DoS attacks. Plus, it incurs delays to get the required tokens because of the latency
In order to solve the Byzantine consensus problem, Satoshi made miners compete to get reward. Meanwhile, he adopted the hash-based POW consensus mechanism to bring about the greatest possible fairness. To create a valid block, a miner uses a different number as the random element of the block header, this number is called the nonce. Depending on the nonce and what else is in the block the hashing function will yield a hash, and then the miner has to find a hash that is below the difficulty target. By doing so, a miner performs certain work, which has no meaning to real life, but keeps the bitcoin network safe and stable.
In the past eight years, bitcoin mining has evolved greatly. Bitcoin Miners graduated from CPU Mining to GPU mining which was followed by the introduction of FPGA Miners and ASICs-based mining hardware. With the massive increase in hashpower, there is a growing demand for electricity. As a result, most bitcoin mining sites in China are set up in power station-dotted southwestern areas and Inner Mongolia.
The current network has a hashrate of 5000PH/S. And the hashrate of per Avaon741, the latest mining machines of Avalon is 8TH/S and its power 1150 watt. Therefore, even if the whole network is equipped with Avaon741, 17 million kwh would be consumed per day. Undeniably, bitcoin mining is a huge waste of energy.
ASIC Miners with customer chipsets are designed for the sole purpose of Bitcoin mining, they serve no other purpose. As such, they allow for miners to save money on electric bills.
As you can see, it had remained a relatively easy job to mine bitcoins from the genesis block to the 260,000th blocks. However, it becomes harder to mine bitcoins when surpassed the 260,000 threshold, which indicates that the network’s overall hashrate exponentially. As of today, the hashrate has grown from 7MH/S to 5000PH/S.
The current bitcoin mining devices have already entered in the era of ASIC, while using ASIC chips to do AI-related basic calculation has just begun.
Deep learning is the most attractive part in the area if AI. In March 2016, Google’s AlphaGo beat Lee Se-dol from South Korea in 4:1. Following that is the updated version of AlphaGo named Master, it competed with 44 chess players from December 2016 to January 2017, and finally gained a record of 60 consecutive wins. Deep learning algorithm can mostly be reflected into basic linear algebra calculus. There are two main features of linear algebra calculus: one is Tensor flow is structured and predictable, the other is its high calculate density. It is these two features that make deep learning perfectly matches with hardware acceleration.
It is important to note that AlphaGo used 170 GPU and 1200 CPU during the competition with Lee Se-dol; while AlphaGo Master just applied a single version of TPU (Tensor Processing UnITs). Thus it can be seen that bitcoin miner and AI deep learning are comparable, both depend on basic chips to process large scale concurrent computing and use ASIC chips are able to greatly improve computation efficiency.
Apart from human VS machine battle, Apple’s intelligent personal assistant Siri also implemented machine learning technology to enhance its ability of understanding and conducting users’ natural languages. Meanwhile, Tesla’s intelligent driving system also improved users’ driving experience (UE) by AI. Also, AI-enabled shopping prediction system from Amazon can give more accurate recommendation to consumers when they are struggling to find the right products.
Thus, the hash calculated by the bitcoin miners who use ASIC devices during the mining process is worthless; while the AI tech based on ASIC chips will benefit mankind a lot.
At present, a bitcoin miner is able to make 7.22 yuan/THash. He is losing money if it costs more for him to run machines non-stop for 24 hours. And the ASIC machines obsolete will become useless.
Bytom introduces matrix operations and convolution operations in the hashing process of mining, making miners friendlier to AI ASICs than GPU and CPU. As a result, the calculation required for blockchain consensus can also be applied to the AI hardware acceleration service, which will generate greater social benefits. On the one hand, the mining market will stimulate the market for artificial intelligence, expanding needs for the depth learning of ASIC chips. On the other hand, outdated miners can be applied to AI hardware acceleration services, saving mining costs to provide a win-win outcome.
In 2016, National development and reform commission said in a statement that China would speed up the development of its AI sector and create a market worth over $15 billion over the next three years. If the rising of cloud computing leads to an increase in demand for server business, then we could predict that when the market for AI reaches over $10 billion, the market for AI ASIC chips will at least surpass $1 billion. If so, by adopting POW algorithm that is friendly to AI ASIC-chips, Bytom could enable miners to be used for AI acceleration services when they are outdated.
I believe that it is already a success if we could make outdated mining machines useful again.