With the breakthrough of algorithms, the improvement of computing power and the accumulation of data, artificial intelligence technology has ushered in a leap-forward development in recent years. Through a full range of commercialization practices, it has gradually changed the original industrial structure, triggered profound changes within various industries, and even changed the way we interact with the outside world. The artificial intelligence of each industry has become an irreversible trend, and this brings about a huge historical opportunity in the field of computing power in the artificial intelligence industry.
Blooming demand for artificial intelligence computing power
The implementation of artificial intelligence algorithms requires strong computing power support, especially the large-scale use of deep learning algorithms, which puts higher demands on computing power. Since 2012, the demand for computing power in artificial intelligence training has continued to grow, doubling every three and a half months, and has grown by more than 300,000 times.
IDC predicts that the investment in computing power in the global artificial intelligence market will exceed $17.6 billion in 2022, accounting for nearly half of the total AI investment. However, the "2018 China AI Computing Power Development Report" pointed out that the main challenge that hinders the development of AI computing is still the bottleneck of computing power. This shows the enormous development potential of artificial intelligence computing resources.
The future of cloud computing combined with fog computing
Currently, the demand for artificial intelligence is reflected in two main directions. One is a centralized hyper-scale data computing center (cloud computing), and artificial intelligence requires a lot of computing power in the algorithm optimization phase. Take Google's AI robot AlphaGo as an example. Before defeating Lee Sedol, the DeepMind team used 48 TPUs. AlphaGo has been trained for more than 30 million games in a few months.
The other direction is based on edge computing (fog computing) in the field of artificial intelligence applications. In the mobile era, a large amount of data in the local storage computing mode can no longer meet the needs of users. Therefore, the computing power will be tilted to the edge with the development of mobile devices and IoT smart devices, showing a trend of distributed deployment. Full coverage is formed like a mobile communication base station, and the separation between the client and the server is achieved.
Thanks to the development of technologies such as blockchain and 5G, many problems faced by fog computing itself will be solved. 5G technology solves the problem of data transmission rate. The blockchain's encryption properties perfectly solve the personal data security problem. Its incentive mechanism can also attract many distributed computing nodes to form a fog computing network that is wide enough to extend to every corner.
With the further development of artificial intelligence research and application, and the coordinated development of the fields of Internet of Things, intelligent manufacturing, and 5G, the computing power network combining cloud computing and fog computing is bound to become the same infrastructure as communications, power, and network.
TuringFog——The infrastructure computing power network driven by blockchain
In the stage of the outbreak of artificial intelligence computing needs, TuringFog（turingfog.ai） has created a unified resource computing platform for resource-intensive and service artificial intelligence industry based on blockchain technology. Make full use of the economic drivers of blockchain and the productivity of new technologies such as 5G to coordinate the interests of all parties involved. Based on the cloud-integrated technology architecture, physical resources such as computing power and perception are shared and shared, providing more efficient and reliable foundation support for the artificial intelligence industry and the fourth industrial revolution.
In the Turing Fog network, any user can become the seller and renter of the computing power. Whether you are providing an idle home computer or a few large computing centers, you can join the Turing Fog Network.
Distributed joint machine learning developed by Turing Fog enables distributed computing to jointly perform algorithmic tasks. At present, the computing performance of each terminal GPU is improved, so that each device is a computing resource that cannot be ignored. With the support of blockchain and 5G technology, Turing Fog can closely integrate distributed computing power, large-scale data and algorithms to create a network of valuable infrastructure.
First proposed CP measuring unit——TU
Because there is no uniform unit of measurement, the computing resources always rely on third-party media in the process of circulation or have been traded in the form of a server entity. It is the first time that Turing Fog defines an objective measurement unit for the production node: Turing Unit (TU). Simply speaking, one TU equals the 24-hour continuous computational power of the GPU GTX 1080 Ti. Thus, the computing resources can directly enter the stage of trading circulation and have a clear value evaluation system.
Turing fog performs TensorFlow's single-precision deep learning training on different mainstream computing platforms. Each GPU trains ResNet-50, ResNet-152, Inception v3, Inception v4, VGG-16, AlexNet and SSD300 on a single GPU, and records the number of pictures per second that can be processed during the training network as a comparison value, so you can easily calculate the computing power units for each GPU.
Fig Turing Fog TFT Point System
The Turing Fog establishes the Turing Fog Token (TFT) and uses the TFT as the native token. The way of using computing power is that users can purchase the power with fiat currency or TFT, and all the funds received will be used to repurchase and burn TFT from the secondary market.
At the same time, 80% of the burned TFT will be released from the portion of production mining, which will be distributed to the miners and cover the platform operation cost. When production mining has been finished, the project side will repurchase TFT from the open market, and half of the TFT will be burned, and the other half will be distributed to the miners. The released or unlocked TFT can flow to the open market, which is “destroy equals to mining”. Under this mechanism, the TFT will always remain in a deflated state.
The nominal total amount of TFT is 1 billion, and the theoretical upper limit is about 650 million pieces, but the actual upper limit is far less than 650 million. After several years of “destroy equals to mining” mechanism, the ultimate limit will be reduced to 100 million. At this time, “destroy equals to mining” plan will stop, and Turing Fog will enter the stage of ecological maturity and stable development.
Value stability system under the futures delivery mechanism
In the Turing fog system, based on the stable computing power unit of TU and futures delivery mechanism, token investors are isolated from AI developers. The price fluctuation of TFT does not affect the purchase behavior of the users. Once TF Token is transferred to TU, the value becomes relatively static and will not be affected by fluctuations in market prices, which can avoid the value system collapse of the computing resources in the later period.
After purchasing the TFT, the user can choose to immediately exchange the corresponding computing resources, or select a certain time to deliver. When the price of TFT rises, the computing power will increase accordingly, and vice versa. Therefore, the price of the computing resources is always begged with its real fiat currency value accordingly. Thus, it not only ensures the user’s demand to keep the value stable but also considers the potential returns on the investment.
Double layer network node and mining rewards
The Turing Fog Network is a double layer network. The bottom layer is the Turing fog production network, which adopts a unified supervision and dispatching system; the upper layer is the Turing fog ledger network, which is an independent blockchain system.
The production network consists of Turing fog production nodes. Turing fog production nodes include cloud nodes and fog nodes. Cloud nodes are computing power clouds built by Turing fog partners. Fog nodes include social idle or partner GPU computing nodes, normal edge computing nodes and other computing devices on the edge of the network.
A total of approximately 700 million TFTs are used to reward resources and services such as computing power, perception and site provided by the Turing Fog Production Node for the Turing Fog Ecosystem. However, due to the “destroy equals to mining” strategy, the final market volume is far less than 700 million. The value of a node is formed by the market mechanism, depending on factors such as the location of the node, the intrinsic value of the data, and the scarcity.
The ledger network is the blockchain system of Turing Fog. It not only bears the ledger responsibility of the production network but also is an independent and complete blockchain system. Using the DPoS consensus mechanism, the number of supernodes is initially 11, and eventually no more than 51. When the miner successfully creates a block, he will receive a certain TFT reward. The total amount of TFT for mining is 5 million, and 500,000 TFT rewards will be released in the first year and followed the strategy of halving every five years.
The generation of supernodes is based on the market contribution of the Turing Fog Ecosystem by large to small sorting in the current period, resulting in a more unbiased and positive ecological environment.
After the completion of the artificial intelligence computing network combining cloud computing and fog computing, Turing Fog will revolve around three main aspects and gradually build an artificial intelligence ecosystem that integrates resources, development, and application:
1．Under the economic incentive mechanism, the resource sharing system will be put into effect, and the infrastructure needs of the AI enterprise, such as computing power, edge, and perception, will be required.
2、Establish a cloud-integrated DApp (Decentralized Application) platform to help develop and deploy innovative AI applications.
3、Design, certificate, and authorized production of Turing fog cloud host (AI training) and fog nodes (AI reasoning) and peripheral equipment.
At present, the Turing Fog team has cooperated with many well-known universities and artificial intelligence companies to provide customers with artificial intelligence computing solutions and has reached a number of strategic cooperation agreements. At the same time, some local governments, artificial intelligence companies, and universities have reached an agreement with the Turing Fog project. These customer resources have laid a solid foundation for future Turing fog sales.
With more cloud nodes and fog nodes joining the Turing Fog network, a larger software and hardware integration platform for artificial intelligence application deployment is jointly established. Turing fog will have both AI reasoning and edge service capabilities, as well as docking the IoT awareness layer to provide data collection and perceptual sharing for AI applications. It not only solves the problem of training and reasoning, but also bring a cloud-to-end, core-to-edge infrastructure platform to the AI industry, thus creating a new computing model, and pushing the field of artificial intelligence into a new stage of development.