The smart computing platform realizes high performance and intelligentization by using the innovative “extra-computation eliminated” computing architecture and “experience search and matching” computing method. By adopting full switched network, it realizes the integration of computing, storage, and communication resources, and eliminates the time consumption of extra computation; through pre-digitization of the function space, to form a countable state set, and adopts neural network to realize search and matching, so as to support intelligent computing.
Simulate human brain for information processing, high-efficiency global optimal searching
Pre-digitization of function space, nonlinear spaces converted into linear set
Can be applied in weather forecast, pharmaceutical R&D, intelligent manufacturing, GPU, BIM, etc
Type | Parameters | Functions | Performance | High-end | Medium-end | Low-end |
High performance processing server | AETHER customized server, eight-core, 16-thread, 8 CPU, 128G memory, 3T hard disk | Data-based module for domain function objects; Support transactional decomposition for scientific computing; Support linear processing for nonlinear spaces; Support data-based processing for function space; Support pre-computation for AI analysis. | 10000+ clusters linear scaling;Ten-thousand computers level dispatching management; Real-time processing performance >20 million transactions per second; Disk read/write history speed >1 million transactions per second. | 12+ sets AETHER customized servers | 3+ sets AETHER customized servers | 1+ sets AETHER customized servers |
Big data search server | Configuration for single node: MBI-6418A-T5H, 4 CPU, 8G memory, 160G hard disk | Computation/storage/communication integration module: by using in-memory computing module, realizes the integration of computing and storage for single node, and eliminates the time loss caused by data migration; by adopting full switched network, realizes the integration of computing and communications between multiple nodes, and eliminates the spatial loss caused by parallel processing. High-density data fast search module: can customize data template according to business scenario; can realize global multi-level search and comparison matching of massive data space through Hopfield neural network, and support for training of high efficient memory reasoning mechanism. | 10000+ clusters linear scaling; Can realize real-time multi-dimensional perspective analysis and business exploration for trillion level data in millisecond level time. | 4 blade boxes = 112 blade servers = 512 computing nodes | 1 blade box = 28 blade servers = 112 computing nodes | 10 blade servers = 40 computing nodes |
Distributed storage system | PB-level scalable rackmount disk storage system | Large-scale associative memory storage module: Supports Hopfield network mode associative memory storage and distributed fault tolerance; can realize plug-and-play of storage units by introducing SDN, and support expansion of storage nodes as needed. | PB-level storage expansion; support online expansion in seconds; storage space can be doubled by using innovative data compression algorithm; support data refreshing in milliseconds. | 1PB disk array | 128T disk array | Optional |
公司目前于成都市双流区在建占地面积约75亩的工业园区,建成后将作为公司技术研发中心、城市轨道交通系统安全保障技术国家工程实验室成都基地,在科研、生产、产业推广等领域发挥关键作用。
实用新型及外观设计专利
发明专利
软件著作权