Intelligent Computing Resource Pool

The intelligent computing power resource pool is a heterogeneous fusion adaptation platform for multi-artificial intelligence computing power, which can realize the effective docking of hardware performance and computing requirements, the effective adaptation of heterogeneous computing power and user needs, and the flexibility of heterogeneous computing power among nodes Scheduling, multi-computing intelligent operation and open sharing. Coordinate processing of various heterogeneous computing power to maximize computing effectiveness, and provide high-performance and highly reliable computing power support for diverse AI application scenarios. The intelligent computing power resource pool consists of four parts: hardware support platform, heterogeneous AI computing power adaptation platform, heterogeneous AI computing power scheduling platform, and intelligent operation open platform. Relying on the fusion architecture of software and hardware, it solves the problems of poor compatibility and low efficiency caused by multiple architectures, and realizes the classification and integration of hardware resources, pooling reconstruction and intelligent allocation through software definition.

Form a software-defined intelligent computing power resource pool through virtualization, enhance the operation capability of the heterogeneous AI computing power operating platform, and optimize the application architecture. The first is to enhance the fine-grained segmentation capability of computing resources. Fine-grained segmentation of computing resources in the intelligent computing power resource pool according to application requirements and business characteristics can maximize the use of computing power, improve resource utilization, reduce computing costs, and avoid the need for large-scale computing equipment clusters. The complicated work of selection and equipment adaptation. The second is the configuration of heterogeneous computing power server chip architecture. It is necessary to configure and set the technology according to the chip architecture of the heterogeneous computing power server itself, so as to further ensure the pooling of heterogeneous computing power resources. At the same time, a virtual resource pool is created on servers, storage, and networks with heterogeneous computing power. The computing power resources required by upper-layer applications are captured in the resource pool through the API interface, and the mapping from the virtual resource pool to the physical resource pool is realized.