With the cloud as the foundation and effective computing power as the core, from the perspective of sustainable operation, a heterogeneous resource pool is built through computing power servers, high-performance storage, and intelligent lossless networks, providing high-performance and high-quality intelligent computing power and technical services for government and enterprise customers, supporting industrial model training, inference and service innovation, and helping the industry to innovate agile and intelligently upgrade.
1. xPU pooling: It provides heterogeneous multi-intelligent computing power, shields the underlying multi-xPU features, provides unified modeling, and establishes training/inference resource pools to give full play to their own unique capabilities.
2. All-flash storage: It solves the performance bottlenecks such as high throughput, low latency, and massive random small I/O in intelligent computing scenarios, and provides high-performance and low-latency distributed multi-form storage such as files, KVs, objects, and blocks to meet the read and write requirements of new workloads.
3. Low-latency, high-speed lossless network: Break through the network bottleneck encountered in distributed training, solve the traditional Ethernet congestion and jitter problems, define low-latency, high-speed lossless network based on RDMA, study the congestion algorithm with fine-grained control, and realize the basic network acceleration of software and hardware collaboration based on xPU, so as to improve the efficiency of distributed training communication.
4. Service-based intelligent computing power: Through cloudification, intelligent scheduling and other means, the flexible use and reuse of computing center resources are realized, the computing power demand is accurately matched, intelligent deployment and intelligent scheduling are realized, and the overall resource utilization rate is improved.
1. computing platform scenario
The intelligent computing cluster based on basic software and hardware has a computing power of million-level PFLOPS FP16, which is equivalent to the computing power of hundreds of thousands of high-performance PCs.
2. Urban intelligent brain platform scenario
Based on Inspur Intelligent Computing Cloud, it builds a smart city cloud base to carry industry applications such as transportation, medical care, government, emergency response, education, environmental protection, and Internet.
3. Scientific research and innovation platform scenario
Build a multi-level and sustainable talent ecosystem, rely on universities and institutes to carry out intelligence teaching cooperation, and strengthen collaborative innovation between enterprises and scientific research institutes.
A regional intelligent computing center
A certain region is rich in scientific and educational resources, attaches great importance to the development of science and technology industry, and has a strong demand for computing power. Relying on the Inspur intelligent computing cloud solution, with the Inspur cloud computing platform ICP as the base and data as the resource, it provides computing power services, data services and algorithm services required for intelligent applications, providing a solid foundation for the development of high-quality industries.