“Computing power is the primary productive force in the digital economy.” On July 29, 2022, at the SenseTime Technology Sub-Forum of China’s First Computing Power Conference, Gao Shanshan, Director of the Shandong Sino-US Digital Media International Cooperation Research Center, said bluntly that artificial intelligence (AI) In this era, computing power infrastructure is constantly changing industries such as finance, medicine, and data centers.
Therefore, AI computing power has become a major increment in the development of the global digital economy and the foundation of the digital economic era.
Computing power represents a new type of productivity. Whoever has the computing power to develop the digital economy industry in the future will have the ultimate power to lead the development of the digital economy industry in the future. In the past, we talked about some computing power and the operating mechanism of information construction, which may focus on the Internet, infrastructure and software systems. and other levels; at this stage, AI computing power applications are transforming from To C (consumer) side to To B (enterprise) side and To G (government) side. The scenarios are more diverse, the data resources required are richer, and the computing power required is more. Outstanding; in the future, under the intensity of high computing power, it will be necessary to have both timely response and personalized data analysis and application capabilities in different industries.
As the demand and competition for AI computing power begins, OpenPower comes into being. OpenPower high-performance distributed intelligent computing network platform was initiated and established by VV GLOBAL Foundation and Oasis Innovation Labs. OpenPower purchases computing power from centralized supercomputing equipment suppliers in 20 countries and regions around the world, and provides stable and low-cost computing power leasing, technology expansion, and analytical data services to many AIGC industrial institutions and AI generation systems. At the same time, with the help of the OpenPower supercomputing protocol and the computing power supply network system, global users can have the opportunity to participate in the core computing power supply chain of the entire AIGC industry and obtain stable and high-yield investment returns in the computing power blue ocean market.
High-end challenges facing AI computing power
the threshold for applying AI computing power is too high ;
Second, there are contradictions in the application of AI computing power in traditional industrial scenarios and insufficient incentive mechanisms ;
In order to solve the above problems and break the computing power crisis, OpenPower launches the challenge starting from technology. First, we must understand how computing power works. Taking AI computing power as an example, CPU, GPU, DSP, etc. can all run computing power , but There is still a dedicated AI computing chip, why? It’s also related to computing power.
– The CPU (central processing unit) is a general-purpose processor that can handle everything. It is like a Swiss Army knife. It can do everything but is not professional and efficient.
– GPU (Graphics Processing Unit) is a processor specially used to process graphics and images. Compared with the CPU, the data type processed by the GPU is single. Because of the computing power and AI computing power and the ease of forming a large cluster, when performing AI operations, performance, It is far superior to CPU in many aspects such as power consumption and is often used to handle AI operations.
– DSP (digital signal processor) is specially used to process digital signals. DSP is similar to GPU and can also be used for AI operations, such as Qualcomm’s mobile phone SoC.
AI computing power chip is a chip specially used to process AI-related calculations. This is different from the “part-time” AI computing of CPU, GPU, and DSP. Even the most efficient GPU has a gap compared with the AI computing power chip. AI Computing power chips comprehensively surpass the various processors mentioned above in terms of latency, performance, power consumption, and energy efficiency ratio. OpenPower uses TPU as a professional chip for AI computing power . The main computing resources of TPU are:
– Matrix Multiply Unit: Matrix multiplication unit
– Accumulators: stores the intermediate results of matrix multiplication and addition outputs
– Activation: activation unit
– Unified Buffer: unified cache
TPU processing speed is 15-30 times faster than GPU and CPU, and in terms of energy efficiency, TPU is improved by 30 to 80 times. This is not surprising, because the CNN operation run by TPU is mainly matrix multiplication. This is the benefit of dedicated chips. .
OpenPower aims to create a larger, lower-cost and more efficient computing power chain network and collection, distribution and scheduling network. Different from traditional centralized computing power networks and cloud computing power networks , OpenPower will create the world’s largest intelligent computing cluster resources. pool.
At present, the space area deployed by global computing power centers is insufficient, data assets are difficult to access, and computing power application content is insufficient. These are the challenges for AI computing power application at this stage. OpenPower will accelerate the construction of artificial intelligence computing centers. According to OpenPower , future computing centers are not defined for a certain industry, enterprise or customer, but focus on global industrial collaboration.
OpenPower ‘s goal is to build the world’s largest high-performance distributed intelligent computing network and lay a solid foundation for AI computing power for another industrial revolution and civilizational leap in human society. As a high-performance distributed computing power supply platform , OpenPower will set up independent operation centers in the United States, Canada, Russia, Australia, Estonia, Indonesia, Thailand , Hong Kong, China and other countries and regions , with community resources almost all over the world, realizing intelligent The large-scale implementation of computing centers will provide efficient green computing power for applications such as artificial intelligence.