Huawei Unveils New AI Storage Solutions for the Era of Large Models
-
China Economic Net, July 14 (Reporter Wang Huimian) — Today, Huawei released new AI storage products tailored for the era of large models, providing optimal storage solutions for foundational model training, industry model training, and scenario-specific model training and inference, unleashing new AI momentum.
Zhou Yuefeng, President of Huawei's Data Storage Product Line, stated: "In the era of large models, data determines the intelligence level of AI. As the carrier of data, data storage has become a critical infrastructure for AI large models. Huawei Data Storage will continue to innovate, offering diversified solutions and products for the AI large model era, collaborating with partners to empower industries with AI."
It is reported that Huawei, aligning with the AI development trends of the large model era, has launched the OceanStor A310 Deep Learning Data Lake Storage and the FusionCube A3000 Training/Inference Hyper-Converged Appliance to cater to large model applications across different industries and scenarios.
The OceanStor A310 Deep Learning Data Lake Storage is designed for foundational and industry-specific large model data lake scenarios, enabling end-to-end AI massive data management from data aggregation and preprocessing to model training and inference applications. The OceanStor A310, in a single 5U chassis, supports the industry's highest bandwidth of 400GB/s and peak performance of 12 million IOPS, with linear scalability up to 4,096 nodes and seamless multi-protocol interoperability. Its Global File System (GFS) enables cross-region intelligent data orchestration, simplifying data aggregation processes. Near-storage computing enhances preprocessing efficiency by 30% by reducing data movement.
The FusionCube A3000 Training/Inference Hyper-Converged Appliance targets industry-specific large model training and inference scenarios, supporting applications with tens of billions of parameters. It integrates OceanStor A300 high-performance storage nodes, training/inference nodes, networking devices, and AI platform and management software, offering a turnkey deployment experience for large model partners with one-stop delivery. It is ready to use out of the box, with deployment completed within two hours. Both training/inference nodes and storage nodes can be independently scaled horizontally to meet varying model demands. Additionally, the FusionCube A3000 utilizes high-performance containers to enable GPU sharing across multiple model training and inference tasks, increasing resource utilization from 40% to over 70%. The appliance supports two flexible business models: Huawei's Ascend one-stop solution and third-party partner solutions with open computing, networking, and AI platform software.