Hewlett Packard Enterprise Co. said today it’s expanding the capabilities of HPE GreenLake for File Storage with the launch of new, high-density all-flash storage options designed to support data lakes and manage artificial intelligence workloads.

The company is making some big claims about the improved performance of its updated offering, saying it’s delivering four times the capacity and two-times the performance per rack unit. With these improvements, HPE says, enterprises will be able to scale AI throughput by two times while significantly reducing power consumption, with energy-efficiency gains of up to 50% in some use cases. In turn, it adds, this will allow organizations to accelerate their AI workloads while reducing power consumption and their overall data center footprint.

HPE GreenLake for File Storage is a part of the HPE GreenLake portfolio of as-a-service offerings, which allows customers to rent on-premises data center equipment and manage it as though it were a cloud service. The offering consists of standardized, composable building blocks made up compute nodes, commodity storage and network switches that can be configured for different needs.

In a blog post, David Yu, senior manager of product marketing for HPE Storage, said the new offering solves three key challenges that are preventing organizations from realizing the potential of AI and achieving greater scale to unlock more value from their data.

Performance boost

Yu said GreenLake for File Storage features some key innovations that make it ideally suited for accelerating and scaling AI workloads, including training, fine-tuning and inference. They’re important, he said, because performance limitations at scale can be a huge hindrance for AI storage, which can scale in terms of capacity but fail to keep up in terms of performance.

“HPE GreenLake for File Storage accelerates your most data-intensive applications with enterprise performance at AI scale,” Yu said. “This is performance that spans all the stages of AI — from data aggregation, data preparation, and training and tuning to inferencing. And it’s not just performance that reaches a peak at some point in time for a small data set. Instead, it’s fast, sustained performance that spans the entire scale of your data for the most demanding, data-intensive AI applications, including gen AI and large language models.”

According to Yu, GreenLake for File Storage achieves this thanks to its disaggregated, share-everything and highly-resilient modular architecture, which allows capacity and performance to be scaled independently of one another. Designed for exabyte-scale workloads, it enables data to be transferred at rapid speeds to ensure fast and predictable performance with no front-end caching. Its architecture means there’s no data movement between media and no tiered data pipelines, making it ideal for boosting the most data-intensive AI applications.

Simplified management

GreenLake for File Storage also solves problems around legacy file storage that make it highly inefficient for AI workloads. Traditionally, managing file storage requires specialized domain expertise and is a tedious, manual process. Yet these management tasks are a necessary evil for organizations looking to scale up AI.

With its cloudlike experience, GreenLake for File Storage simplifies file data management tasks to reduce the burden on information technology teams. Yu reeled off a whole list of benefits, including streamlined deployment, simple setup, easy file share creation, fast job completion and unified storage management through a single cloud console that can be accessed on any device in any location.

According to Yu, these capabilities provide real-world advantages: “Your data scientists and line-of-business application owners are no longer burdened by cumbersome, legacy file systems that require technical expertise and intricate setup processes just to run,” he said.

Improved scalability

A third challenge with traditional file storage relates to AI’s insatiable appetite for storage capacity, rack space and power consumption. With legacy storage systems, it’s simply impractical to scale AI workloads beyond a certain point without building an entirely new data center. “Legacy NAS solutions with shared-nothing architectures are unable to efficiently scale out to keep pace with the capacity density, cost per terabyte and power efficiency demands of AI workloads,” Yu explained.

GreenLake for File Storage features various innovations to solve this challenge, enabling organizations to reduce their storage costs by up to four times and power consumption by two times. It combines advances such as no-overhead snapshots and native data replication with the superior efficiency of flash storage and the company’s proprietary Similarity data compression algorithm to enable more effective data reduction.

In addition, Yu said, GreenLake for File Storage now supports optimized utilization of graphics processing units through Nvidia Corp.’s InfiniBand, GPUDirect and RDMA networking technologies. This enables faster checkpointing in order to accelerate AI workloads such as model training and tuning, he said.

“To power your AI initiatives and compete in today’s marketplace, you need an AI-ready file storage solution that can deliver enterprise performance, simplicity, and enhanced efficiency,” Yu said. “Truly accelerating your AI workloads requires all three measures of AI scalability, as each one is essential for AI success.”

Images: HPE