HangzhouMay 28, 2026 /PRNewswire/ — Currently, AI large models have reached a scale of trillions of parameters and millions of Token context lengths, with the scale continuing to increase. Intelligent computing centers have entered the “system efficiency competition” phase. To meet the training and inference needs of AI large models, efficient operation of GPU clusters must rely on the overall coordination of computing power, network, storage, scheduling, and operations. Among these, the storage system is evolving from a traditional data-bearing platform into a critical infrastructure that impacts GPU utilization, training efficiency, and business continuity.
Facing demands such as massive training data, high-concurrency access, fast Checkpoint reads and writes, and mixed cold and hot data management, Cloudtip Information has launched a high-performance, highly reliable, and scalable distributed storage solution.
Cloudtip Information Distributed Storage Solution
The Cloudtip Information Distributed Storage Solution adopts an architecture design of “high-performance all-flash layer + large-capacity hybrid flash layer + high-speed lossless network + distributed parallel file system + multi-protocol unified access.” Leveraging the synergy between the CS8000 X6 and CS6000 X6 dual products, it builds a layered data foundation, fully adapting to large model training and inference scenarios as well as high-performance computing scenarios, making it a preferred solution for intelligent computing center construction.
Dual Product Synergy: CS8000+CS6000 Builds a Layered Data Foundation
In the layered architecture, the CS8000 X6 addresses high-performance data access needs, with a single-node sequential read bandwidth of up to 40GB/s; in typical small file test scenarios, IOPS can reach 400K+. As the all-flash acceleration layer, it primarily handles demands such as large model training hot datasets, high-frequency Checkpoint reads and writes, model loading and distribution, high-concurrency file access, and low-latency data access.
The CS6000 X6 addresses massive data storage needs, supporting large-scale horizontal expansion with a single file system capacity of over 200PB. As the hybrid flash capacity layer, it primarily handles demands such as massive unstructured data, mixed cold and hot datasets, full training data and historical data, log archiving and backup, large-capacity files like video images and documents, and medium- to long-term data lifecycle management.
The synergy between the two forms a distributed storage system of “high-performance access + large-capacity storage,” meeting data access and long-term management needs in AI training, cluster inference, and HPC scenarios.
From Technical Capabilities to Business Benefits, Solving Industry Pain Points Across All Dimensions
Based on the above architecture, the Cloudtip Information Distributed Storage Solution supports diverse data access needs in AI training, cluster inference, and HPC scenarios from dimensions such as performance release, data protection, elastic scaling, protocol compatibility, smooth migration, and intelligent management.
1. High-Performance Parallel Access, Boosting GPU Cluster Utilization
AI training, HPC simulation, and multimodal data processing require high throughput, low latency, and high concurrency access. The solution leverages high-performance distributed parallel storage in coordination with high-speed networks to reduce I/O bottlenecks common in large-scale concurrent access scenarios, minimizing GPU wait times for data and enabling more efficient use of computing resources.
2. Layered Data Storage, Optimizing Storage Costs and Data Tier Management
Facing the continuous growth of training data, model files, Checkpoints, intermediate results, logs, and archived data, the solution uses the CS8000 all-flash high-performance layer and the CS6000 hybrid flash capacity layer to tier high-frequency hot data and low-frequency cold data, ensuring performance while reducing the cost pressure of full all-flash deployment.
3. Multi-Protocol Unified Access, Reducing Data Silos and Migration Complexity
The solution adapts to AI, HPC, big data, and general business environments through multi-protocol compatibility and unified data access capabilities, reducing data duplication, migration complexity, and management fragmentation caused by multiple storage systems, effectively lowering data migration costs.
4. Distributed High-Reliability Architecture, Supporting Business Continuity and Elastic Scaling
As intelligent computing centers evolve from tens or hundreds of cards to thousands or tens of thousands of cards, storage must not only expand capacity but also scale performance synchronously. The solution supports business continuity under disk, node, and network failures through multi-replication, erasure coding, snapshots, remote replication, automatic fault repair, and horizontal scaling capabilities, while also supporting future scale expansion of the intelligent computing center.
Supporting Next-Generation Intelligent Computing Centers with an Efficient Data Foundation
In the construction of intelligent computing centers, computing power determines the upper limit, storage determines efficiency, and network determines coordination. The Cloudtip Information Distributed Storage Solution, centered on the CS8000 X6 and CS6000 X6, builds a layered storage system covering high-performance hot data access and large-capacity data management, providing stable support for AI training, cluster inference, HPC, and massive unstructured data processing.
In the future, Cloudtip Information will continue to enhance its server, storage, network, liquid cooling, and operations service capabilities for AIDC intelligent computing centers, supporting customer intelligent computing infrastructure construction and continuous business evolution with an efficient, reliable, and scalable data foundation.