Shared todayAI seriesIn-depth Research Report:AI computing power: "PARFAITE" computing power opens a new era of computing
Report produced by KPMG).
Report total: 78 pages.
Featured Report**: The School of Artificial Intelligence
History is repeating itself, and technology is always moving forward. Looking at the past three scientific and technological revolutions, all of them are driven by some basic general purpose technologies (GPTS), standing at the moment of the outbreak of ChatGPT, general artificial intelligence is very likely to become a new generation of GPTS. From GPT (generative pre-trained models) to GPTS, the fourth technological revolution is coming, and artificial intelligence will also usher in the fourth wave of development. Under the background of the superposition of two waves, the digital voyage of mankind has opened a new chapter. As a new type of productivity, computing power is the underlying driving force for economic and social development, and the contradiction between its limited supply scale and exponential growth in demand is becoming increasingly prominent, the traditional computing paradigm is failing, and the computing mode is evolving from central to distributed, and then to the cloud computing stage, and is about to enter the stage of integrated intelligent computing, marking the beginning of a new computing era. The competition for computing power dominated by large countries and large enterprises will intensify, and capital, technology, and talents are indispensable. The most fundamental thing is to accurately grasp the direction of force in order to win the future.
In the future, the development of computing power will have two core characteristics: "the infrastructure of the digital economy" and "the core power of general artificial intelligence". Correspondingly, computing power will accelerate in two key dimensions: inclusive and lntelligent. "Universal" means that computing power will become a "3A" basic resource that is affordable, ailable, and adaptable, and is a solid foundation for the development of the digital economy. "Intelligence" means that computing power will have the "3S" intelligence represented by self-adaptation, self-learning and self-evolution, which is an inevitable requirement for realizing the wide application of general artificial intelligence and promoting the digital economy to a higher stage.
The outbreak of the previous scientific and technological revolutions of mankind has been marked by the leapfrog development of productive forces caused by a new generation of general purpose technologies (GPTS), which in turn has led to subversive changes in the means of production and economic paradigms.
Different from SPTS (Special Purpose Technologies), GPTS represents a revolutionary technology that has a huge, far-reaching and extensive impact on human economy and society, typical technologies include wheels, farming technology, steam engines, water, coal and electricity, computers, Internet, etc.
In the era of agricultural economy, the popularization of wheels and farming techniques has improved the efficiency of people's migration and access to food, and human beings can continue to reproduce on the basis of large-scale exploitation of land resources, providing a steady stream of underlying power for economic development. However, due to the low efficiency of pure manual labor and limited land resources (especially arable land is a non-renewable resource), the growth of agricultural economy is relatively slow and there is an obvious "ceiling".
In the era of industrial economy, the steam engine and electrification triggered the first industrial revolution and the second industrial revolution respectively, these two GPTS fundamentally changed the nature of manual labor, replaced by mechanized mass production, driven by new productive forces such as steam power and electricity, the efficiency of human access to natural resources such as minerals has been greatly improved, and the rise of modern factories characterized by large-scale assembly line production has accelerated, promoting the leapfrog development of human society.
In the era of digital economy, information and communication technology (ICT), as the GPTS that promotes the information technology revolution, has fundamentally changed certain types of mental work, greatly improved the efficiency of human access to information and processing information, and the digital economic model with computing power as productivity and information and data as means of production has begun to sprout.
At the end of November 2022, Openal's ChatGPT (Chat Generative Pre-Trained Transformer) sparked heated discussions around the world, and artificial intelligence technology is expected to become a new generation of GPTS - Artificia General Intelligence (AGI). AGI will promote the digital economy to flourish more vigorously and bring the development of human society to a higher stage, from GPT (generative pre-trained model) to GPTS (general technology), artificial intelligence is expected to be a new underlying general technology, igniting the fourth scientific and technological revolution!
AGI is also known as strong artificial intelligence (strong al), which refers to machine intelligence that is connected with human characteristics such as consciousness, perceptuality, knowledge, and consciousness, and can perform human intellectual behavior. Correspondingly, applied AI, also known as weak AI, refers to machine intelligence that can implement specialized or specific skills, such as facial recognition and machine translation.
So far, the various artificial intelligence systems that humans have come into contact with belong to applied AI, which can surpass humans in some single intelligence, such as playing Go. But if human intelligence is to be reached or even surpassed in a broader sense, AGI must be achieved. However, the development of artificial intelligence in the past has not been smooth sailing, and the future for AGI will not be achieved overnight.
In the context of the superposition of the fourth scientific and technological revolution and the double wave of AGI, a new round of scientific and technological innovation cycle will start with the innovation of underlying computing power, prosper in the explosion of upper-layer applications such as AIGC, and digital technologies such as big data, Internet of Things, artificial intelligence, and blockchain tend to evolve in combination, which will release a huge multiplier effect. While the demand for computing power is growing exponentially, the traditional computing architecture is losing its competitiveness, and a new computing era that breaks the mindset and fully releases the value of the underlying power of computing power is about to open.
Specifically, the scientific and technological innovation cycle generally goes through three stages: "technological innovation-model innovation-application innovation". Taking the popularization of personal computers and mobile Internet as examples, both of them revolve around ICT hardware innovation, which in turn brings business model innovation such as portals and platforms, and finally the continuous emergence of various software applications. Corresponding to the industrial level, the focus of competition will gradually transition from hard technology strength to soft power such as industrial ecological construction.
With the growth of PC and intelligent hardware penetration peaking, the Internet traffic dividend fading, scientific and technological innovation is about to open a new round of cycle, all kinds of cutting-edge technologies are moving towards deep integration, and technological innovation led by hard technology strength is back in the spotlight again.
Computing power in the narrow sense can be measured in terms of the number of floating-point operations per second (FLOPS) that a computer theoretically has3. For example, the world's first general-purpose computer, "ENIAC", can perform only 5,000 calculations per second, and the world's number one supercomputer, "Frontier", can perform exascale operations per second (1194exaflop) 。In the process, the role of computing power in economic and social development is becoming more and more important, and the connotation of computing power has gradually been freed from the limitation of physical carriers and expanded into a new type of productivity integrating information computing power, data storage power and network carrying capacity in the digital economy era.
The evolution of the computing mode in the first three stages is mainly supported by Moore's Law and von Neumann computing architecture, the former continues to guide the chip process improvement, and the latter realizes the separation of computer hardware design and software design, which greatly improves the computing flexibility and forms a general computing system with "CPU (Central ProcessingUnit, ** processor)" as the core. However, Moore's Law is constantly approaching the physical limit, relying on advanced processes to improve computing performance is becoming more and more unsustainable, and more diverse and personalized computing needs are forcing the general computing architecture to innovate in the direction of general-purpose parallelism, improving specificity and realizing customization, such as CPU + GPU, CPU + ASIC, CPU + FPGA and other innovative models.
One clear sign is that the industry is accelerating innovation in chips and computing architectures, and CPUs are losing their dominance. For example, since 2016, Google has been innovating a dedicated chip TPU (Tensor Processing Unit) tailored for machine learning, and has used TPU for a lot of AI training work. In addition, there are also views that TPU and GPU are not the optimal solution for general artificial intelligence, pointing out that quantum computing has powerful parallel computing capabilities that far exceed classical computing in principle, IBM announced in 2023 that it will cooperate with the University of Tokyo and the University of Chicago to build a quantum computer driven by 100,000 qubits (the basic unit of quantum information processing), which is expected to promote quantum computing in the research and development of new drugs. Explore applications for dark matter, code-breaking, and more.
All in all, in the stage of integrated intelligent computing, computing power, as the "underlying driving force", has a more extensive and profound impact on economic and social development, and computing power is shifting from the "passive" development driven by the demand for terminal computing to the "active" development represented by promoting AI large model training and realizing general artificial intelligence beyond classical computing. New hardware and new architectures are emerging, and existing chips, operating systems, and application software may be overturned and rebuilt, and a new computing era is about to emerge.
Whether passive or active, they all point to the same conclusion, the growth rate of computing power supply is obviously difficult to meet the exponential explosion of demand, and the reserve of computing power has become a necessary move for all walks of life. Computing power has the characteristics of high investment, high risk, strong externality and high monopoly, which means that the competition around computing power will be dominated by the participation of large countries and large enterprises. Facing the next round of competition, hard-core technical strength, strong financial reserves, and high-level talent teams are all "plus points" to participate in the next round of competition, and only by accurately grasping the direction of force is the "must" to win the future.
Report total: 78 pages.
Featured Report**: The School of Artificial Intelligence