How to solve the "tidal idleness" problem of urban computing power? Guixu Laboratory has proposed a "high-dimensional scheduling" solution to seek pilot verification from governments and enterprises.

Double 11-Dimension Global Scheduling System Optimizes Urban Computing Power Network, Seeks Government-Enterprise Pilot Cooperation

With the rapid implementation of large AI models, urban digital twin and artificial general intelligence technologies, the computing power demand of the whole society presents explosive, dynamic and large-scale growth, and the drawbacks of the traditional decentralized computing power construction mode continue to become prominent. At present, there are common industry pain points in domestic urban computing power construction, such as isolated resources, static scheduling, extensive energy consumption control, and difficult cross-subject circulation. A large number of intelligent computing centers and enterprise private computing power clusters have tidal idle problems, and the comprehensive utilization efficiency of computing power is relatively low, which is difficult to support high-load new computing power scenarios such as continuous deduction of world models, long time-series simulation, and operation of urban global digital twins. In response to the industry's rigid demand for the upgrading of urban computing power infrastructure, the Guxu Theory Laboratory relies on its self-developed Yi-li equilibrium logic and the Double 11-Dimension Global Unified Large Model System to build a city-level global computing power scheduling solution, realizing that computing power resources are schedulable, metered, and available on demand just like power resources. At present, the project is open to government-enterprise pilot cooperation for key cities.

Targeting the Core Pain Points of the Computing Power Industry to Fill the Gap in Urban Computing Power Coordination

At this stage, the construction of computing power networks in domestic first-tier cities has formed a large-scale hardware foundation. The computing power highlands represented by Shanghai have gathered diverse computing power resources such as supercomputing centers, municipal intelligent computing centers, private GPU clusters of leading enterprises, and edge computing power nodes, with hardware computing power reserves ranking among the top in the country. However, at the level of overall computing power scheduling, there are still obvious shortcomings in the industry, which cannot match the infrastructure requirements of the AGI and world model era.

At present, various types of urban computing power resources are in a state of mutual fragmentation. State-owned public computing power, enterprise private computing power, and scientific research dedicated supercomputers are managed by different subjects, and there is a unified global collaborative scheduling mechanism. A large number of computing power resources are idle for a long time during the flat peak period, while computing power shortage occurs during the peak period, and the tidal mismatch between supply and demand is prominent. Traditional computing power scheduling mostly adopts a static reservation mode, which is only adapted to traditional AI tasks such as batch training and offline inference, and cannot elastically allocate real-time fluctuating loads such as continuous iteration of world models and urban dynamic twin simulation.

At the same time, resources such as land, power and energy consumption indicators in core urban areas are scarce, and the expansion space of computing power hardware is limited. The expansion mode that purely relies on adding new computing power hardware is costly and extremely difficult to implement. The industry urgently needs a global computing power governance system that does not add new hardware and only optimizes scheduling, which can systematically upgrade urban computing power infrastructure by revitalizing existing computing power, optimizing resource allocation and reducing comprehensive energy consumption. Traditional computing power management systems can only realize single-cluster load monitoring, lack of multi-constraint, high-dimensional, global intelligent optimization capabilities, and the market has long lacked an integrated computing power base solution adapted to megacities.

Build a High-Dimensional Computing Power Scheduling System to Realize Universal and Accessible Supply of Computing Power Like Electric Power

Different from the traditional single-dimension computing power monitoring and scheduling tools, the Guxu Theory Laboratory innovatively integrates the underlying logic of Yin-Yang balance in Yi-li and the self-developed Double 11-Dimension Global Digital Integration System, builds a high-dimensional simulation scheduling system covering all elements of computing power, reconstructs the operation paradigm of urban computing power infrastructure, and promotes the upgrading of computing power from fragmented resources to urban public inclusive infrastructure.

The project defines the operation system of computing power corresponding to the Yin-Yang balance law in Yi-li: computing power deduction operation is defined as dynamic Yang energy, while storage, network, scheduling, security and energy consumption constraints are defined as static Yin carriers. Through the dynamic balance principle of Yin and Yang, the core problem of the traditional computing power system where Yin and Yang are separated and resources cannot circulate and store energy is solved. Relying on the Double 11-Dimension system, the system incorporates 11 core elements: computing power, storage capacity, network bandwidth, energy consumption indicators, node location, task type, security isolation, elastic quota, fault redundancy, load time sequence, and computing power pricing, converts all urban computing power nodes into standardized state vectors in a high-dimensional space, and realizes a panoramic digital portrait of global computing power.

Based on the global high-dimensional space optimization algorithm, the system can identify surplus nodes and gap nodes of urban computing power in real time, follow the system balance logic of "reducing the surplus and supplementing the deficiency", automatically complete cross-region and cross-subject computing power task migration and resource allocation, and realize sub-second dynamic elastic scheduling. At the same time, the system builds a computing power energy storage buffer mechanism, combined with the new computing-in-memory storage technology to cache intermediate inference data of AI models, which realizes peak shaving and valley filling of computing power and greatly smooths the supply and demand fluctuation of urban computing power.

Taking the Shanghai computing power network as a simulation pilot, the system can increase the utilization rate of urban computing power resources from the original 35%-48% to 72%-81%, and significantly optimize the PUE value of unit computing power, effectively alleviating the pressure of urban energy consumption control. The whole solution does not need to transform the existing GPU and intelligent computing center hardware equipment, and can complete the systematic quality and efficiency improvement of urban computing power infrastructure through the upgrade of upper-layer algorithms and scheduling systems, which can be fully adapted to new computing power scenarios such as large model training, AGI long-range reasoning, urban world model simulation, and dynamic operation of digital twins.

Implement the Commercial Model of Government-Enterprise Cooperation to Open Up the Broad Market of Urban Computing Power Upgrade

The core service targets of this project are big data centers, development and reform commissions, economic and information departments, state-owned computing power platforms and various science and innovation park operation entities in various cities, focusing on four core services: global urban computing power coordinated governance, existing computing power revitalization and optimization, coordinated control of computing power energy consumption, and AI computing power scenario adaptation and upgrading.

Different from the one-time system delivery mode of traditional technology projects, the project adopts a long-term business model of "pilot deployment + continuous iteration + operation and maintenance services". By implementing the urban computing power scheduling brain, it provides the government with an integrated solution of global computing power monitoring, simulation deduction, resource coordination, energy consumption optimization, and computing power asset metering. At the same time, relying on the standardized computing power metering and circulation system, a compliant computing power sharing and circulation mechanism is built to help cities build a market-oriented allocation system for computing power elements.

At present, major cities across the country have started the construction and upgrading of new computing power infrastructure. The revitalization of existing computing power, energy conservation and carbon reduction of computing power, and global computing power coordination have become the core rigid demands of digital city construction, and the trillion-level computing power infrastructure optimization market continues to release dividends. With the unique high-dimensional global scheduling theoretical system and city-level simulation implementation capability, this project is different from traditional single-dimension computing power scheduling products, and has extremely strong technical scarcity and scenario adaptability.

At this stage, the project has completed the construction of a complete theoretical system, the iteration of algorithm models and the global simulation verification of the Shanghai urban computing power network, and the technical solution is mature and implementable. In the future, it can be quickly replicated and promoted to first-tier and new first-tier cities across the country, continuously improve the domestic urban public computing power base through technology empowerment, and provide stable, low-carbon and inclusive new infrastructure support for the industrial implementation of AGI and world models.