NEC Selects DDN Storage

NEC Corporation has selected DDN (DataDirect Networks, Inc.) storage for its new HPC dedicated to AI research.

With a raw compute performance of 580 PFLOPS, the new research HPC will provide data scientists at NEC with access to the largest AI-focused development environment in the industry in Japan and assist in the rapid development of advanced AI.

The AI research HPC relies on the DDN EXAScaler parallel file system, which has been used in HPC environments around the world for more than 20 years, and more recently to support emerging large-scale AI use cases.

The storage environment consists of multiple DDN ES400NVX EXAScaler appliances with a total capacity of 16PB, providing 400GB /s read and 320GB/s write IO performance, and around 24 million IO/s. The AI research HPC’s hardware system is configured with NVIDIA’s A100 80GB Tensor Core GPU for processing and NVIDIA Spectrum SN3700 switches for networking, which are integrated with the DDN solutions.

“NEC has been developing various advanced AI technologies such as biometric authentication, image recognition, video analysis, and data analytics. Therefore, we handle a variety of large-scale data, including images, video, audio, and text,” said Takatoshi Kitano, senior AI platform architect, digital technology development laboratory, NEC. “After a year of technical verification and operation, we decided to adopt DDN’s high-performance storage, which can handle a variety of I/O workloads. Another deciding factor in our decision was DDN Japan’s excellent technical development and maintenance capabilities. We expect both companies to collaborate on the development of the AI HPC and take on the challenge of creating new social value in the future.“

In AI, the “data pre-processing” phase-from data ingestion to data transfer to training-is the most critical phase for a storage environment. EXAScaler appliances provide performing storage for a variety of workloads and also offer the security and management features that are essential for enterprises’ AI workloads. This storage system provides GPU direct storage integration, the CSI driver necessary for storage access through Kubernetes, and multi-tenancy-standard features that enable secure and simplified management without compromising on performance improvements when utilized in a deep learning environment.

“Data has become a strategic tool to address a variety of business opportunities,” said Robert Triendl, GM, DDN Japan. “We are excited by this opportunity to supply NEC with an efficient solution to help drive their data-centric AI initiatives.“

DDN’s storage solutions will support the AI research HPC to realize NEC’s objective of “AI R&D to accelerate the digital transformation of society.” The NEC HPC for AI research has already entered production and is expected to reach its planned capacity in March 2023.