Interview with Deng Zian: How Novus Enhances GPU Utilization through Metadata and Data Path Separation

With the rapid development of artificial intelligence (AI) technology, the demand for data infrastructure among enterprises is increasing, especially in the context of operating large-scale GPU environments. NetApp's recently launched NetApp Novus™, the world's fastest storage system, is designed to meet this demand. To gain insights into NetApp's long-term strategy in the AI infrastructure space and the unique advantages of its new product, the Techritual editorial team interviewed Deng Zian, General Manager of NetApp Hong Kong, Macau, and Mongolia.

AI-Driven Data Infrastructure

Deng Zian pointed out that NetApp's long-term strategy is to assist enterprises in transforming data into AI-ready data, ensuring that data can be accessed and utilized by AI, regardless of whether it is located on-premises, at the edge, or in the cloud. He emphasized that when driving AI applications, enterprises must consider not only the performance of computing resources and infrastructure but also how to discover, prepare, protect, and govern the data that drives AI applications.

"AI is driven by data, and data is driven by NetApp."

Deng Zian, General Manager of NetApp Hong Kong, Macau, and Mongolia

He mentioned that the recently launched NetApp AI Data Engine (AIDE) aims to assist enterprises in discovering and governing data across different environments, enabling better application of this data in AI.

Unique Advantages of NetApp Novus

When discussing the core technology of NetApp Novus, Deng Zian emphasized that this storage system is designed to address the challenges of large-scale AI, featuring a unique architecture that can achieve a total throughput of up to 100TB/s within a single namespace and support hundreds of thousands of GPUs. He noted that this design allows enterprises to meet the demands of AI factories without needing to rebuild their entire infrastructure.

"The unique advantage of Novus lies in integrating industry-leading performance, scalability to handle high AI workloads, and enterprise-grade data management into a single architecture."

Deng Zian, General Manager of NetApp Hong Kong, Macau, and Mongolia

This integration capability not only enhances performance but also allows enterprises to maintain efficiency, security, and operational simplicity when scaling AI infrastructure, avoiding additional system complexity issues.

Addressing the Challenges of AI Factories

Deng Zian discussed the challenges faced in designing NetApp Novus, emphasizing that the operational scale of AI factories is vastly different from traditional enterprise environments. Enterprises need to support multiple workloads simultaneously, including training and inference, while ensuring that the GPU environment receives a stable data supply. He pointed out that these considerations led NetApp to base the design of Novus on NetApp ONTAP and adopt open standards like pNFS, allowing enterprises to continue leveraging their trusted data infrastructure.

"Our goal is not only to enhance performance but also to provide enterprises with the scalability required for AI."

Deng Zian, General Manager of NetApp Hong Kong, Macau, and Mongolia

This design not only improves performance but also allows enterprises to remain flexible in the face of changing market demands.

Key to Enhancing GPU Utilization

Deng Zian explained that GPU utilization depends not only on computational power but also on the performance of the storage system. If the storage system fails to deliver data promptly, the GPU must wait, impacting computational efficiency. The design of NetApp Novus aims to reduce such bottlenecks by separating metadata services from data transmission paths, allowing for independent scaling and ensuring that data can be efficiently transmitted to GPUs.

"Novus helps enterprises make more effective use of their computational resources by separating metadata services from the actual data transmission paths."

Deng Zian, General Manager of NetApp Hong Kong, Macau, and Mongolia

This design not only enhances GPU operational efficiency but also enables enterprises to achieve higher returns in AI environments.

Demand for AI Infrastructure in Hong Kong

Deng Zian noted that the demand for AI infrastructure in Hong Kong is rapidly growing, primarily due to government investment in public resources and enterprises' adoption of AI. He mentioned that Hong Kong's overall computing power has reached 5,000 petaFLOPS, and the government's "AI Funding Scheme" has approved multiple R&D projects, indicating strong market demand for AI.

"The Hong Kong market is moving from the AI trial phase to broader applications."

Deng Zian, General Manager of NetApp Hong Kong, Macau, and Mongolia

He emphasized that as enterprises' demand for AI workloads increases, NetApp will focus on helping enterprises build data infrastructures capable of supporting high-performance AI environments.

Future Directions of Storage Technology Development

Deng Zian believes that in the coming years, storage technology will no longer just be a place for data storage but will become an essential component of enterprise intelligence layers. He pointed out that as AI becomes more deeply integrated into business operations, data preparation and management will become critical. He anticipates that the market demand for flexibility will increase, as enterprises seek to enhance performance and expand capacity based on workload requirements.

"The future of storage technology is not just about storing more data but making data a trusted, readily available resource."

Deng Zian, General Manager of NetApp Hong Kong, Macau, and Mongolia

He stressed that enterprises need to be able to respond flexibly to ever-changing business demands while maintaining efficient operations in an increasingly complex technological environment.

In summary, Deng Zian emphasized NetApp's forward-looking strategy in the AI infrastructure space and anticipated future developments in storage technology, aiming to provide enterprises with more comprehensive data management solutions to support their AI transformation journey.