Getty Images/iStockphoto
On-premises storage, infrastructure evolves to keep pace with AI
Pure Storage is the latest infrastructure vendor to add Nvidia DGX SuperPod certification and new product offerings to support generative AI workloads.
- Scott Sinclair,
-
Enterprise Strategy Group
We provide market insights, research and advisory, and technical validations for tech buyers.
The era of artificial intelligence is upon us, and infrastructure providers have responded by updating their systems to support generative AI workloads.
In an August 2023 research study conducted by TechTarget's Enterprise Strategy Group, 54% of respondents said they expect to have a generative AI initiative in place in the next 12 months.
With AI, like many new or emerging workloads that require a level of infrastructure modernization, the initial expectation is that many organizations will turn to public cloud services.
The value that AI delivers, however, is tied to the quality of the data it utilizes. For the modern enterprise, data is everywhere: in the cloud, on premises and at the edge. As the exuberance surrounding the business potential of AI starts to stabilize, organizations are shifting their focus to data security, privacy and locality considerations.
In other words, organizations want to make AI safe before they focus on making it valuable. As concerns about the risk of data use increase, organizations increasingly incorporate on-premises infrastructure into the AI conversation. As a result, the likely future infrastructure for AI-enabled workloads will be a hybrid environment spanning on- and off-premises environments.
Recently, my colleague published a column on the importance of rethinking enterprise storage in the AI era. It was based on several announcements from a recent Pure Storage event, Pure Accelerate, covering the company's platform approach and its recent enhancements to Pure Fusion.
A key capability for hybrid cloud AI success is consistency. The AI era will require organizations to be more efficient in how they identify, manage and use data across location boundaries. On the infrastructure side, that means consistency of management, consistency of performance, and consistency of experience for both IT administrators and developer teams building AI-infused workloads.
The bottom line is that there is not enough expertise in AI to go around, and consistency is one of the best ways of reducing the operational burden of running application environments that span multiple systems and locations. Pure Fusion and Pure Storage's platform approach help to achieve that consistency.
Pure's AI infrastructure updates
In addition to the Pure Fusion announcements, Pure Storage also shared plans to gain Nvidia DGX SuperPod certification by the end of 2024. Certification has quickly become a must-have requirement for AI training environments, as the certification helps verify that the storage can provide the high throughput and low latency necessary for efficient data access and processing.
Pure also introduced a new storage-as-a-service offering, Evergreen One, designed for AI. As part of this offering, Pure Storage guarantees a level of storage performance for GPUs to support AI training and inference workloads. When modernizing the infrastructure to optimize for AI, maximizing GPU utilization is essential to controlling cost. This new service is designed to offer greater consistency and predictability in storage latency for AI workloads to help control the cost of AI infrastructure deployments.
The company also shared plans for integration of secure multi-tenancy at the storage layer. This capability enables the combination of Kubernetes container management with secure multi-tenancy and policy governance to better control and enable data integrations in AI environments. Pure Storage's intent is to help make the storage infrastructure consistent and transparent to the application owners.
Preparing infrastructure for AI
With AI and generative AI poised to transform every facet of business, nearly every IT infrastructure and storage provider is looking to better serve enterprises for AI workloads. In addition to Pure Storage, multiple storage providers have announced plans for Nvidia DGX SuperPod certification, including DDN, Dell, NetApp, IBM, Weka and Vast Data.
As organizations mature in their use of AI workloads, however, simply providing sufficient performance will not be enough. There will be an increased focus on reducing the operational burden of managing and maintaining the infrastructure, the data, the models and the apps.
Consistency -- or ideally, transparency of experience -- will become increasingly valuable. To this end, Pure Storage's Fusion enhancements, its Evergreen//One for AI and its integration of secure multi-tenancy all are examples of the level of consistency and transparency that on-premises infrastructure will need to deliver to be able to provide a modern infrastructure stack comparable with public cloud alternatives. In other words, the infrastructure must simplify not just IT's job, but the work of the application developers and owners as well.
AI requirements, models and data sets are continually evolving and will require systems that can be tailored easily to the needs of specific use cases. In a crowded field of storage providers for AI, Pure Storage highlights how enterprise storage and on-premises infrastructure need to evolve to compete in the era of AI.
Scott Sinclair is Practice Director with TechTarget's Enterprise Strategy Group, covering the storage industry.
Enterprise Strategy Group is a division of TechTarget. Its analysts have business relationships with technology vendors.
