Pure Storage, Nvidia unveil new AI-ready infrastructure

By on
Pure Storage, Nvidia unveil new AI-ready infrastructure

Flash storage developer Pure Storage together with technology partner Nvidia Wednesday unveiled the next generation of their AIRI AI-ready infrastructure that combines Pure Storage’s new FlashBlade//S: array with Nvidia’s DGX GPU and promised to update that infrastructure to more advanced Nvidia GPUs as they become available.

The new AI-ready architecture was introduced at the Pure//Accelerate Digital techfest22 conference being held this week in Los Angeles.

The new AIRI//S, which includes the FlashBlade//S: and Nvidia’s DGX A100 systems, provides a simple on-demand infrastructure aimed at accelerating AI initiatives, said Amy Fowler, vice president of product marketing for FlashBlade at Pure Storage.

The AIRI was first introduced in 2018 with Nvidia’s DGX-1 and were followed in 2019 with an upgrade to the Nvidia DGX-2 and in 2020 with the A100, Fowler told CRN US.

“And as Nvidia continues to introduce next generations of its DGX, we will continue to further develop the AIRI//S,” she said.

When it was first introduced, AIRI was the world’s first infrastructure for building AI-based infrastructures, reducing 50 full racks of equipment to 50 rack units of space, said Tony Paikeday, senior director of AI systems for Nvidia.

“Customers used to think AI was out of reach, but they don’t think that way now,” Paikeday told CRN.

AIRI is already in use with the world’s largest social media company’s recommendation system, as well as for language processing, pathology, radiology and autonomous driving systems, Paikeday said.

Traditionally, customers needed discrete server clusters to handle data analytics for AI development, clusters for model training and clusters for inference, he said.

“These were all different,” he said. “Each is optimized to deliver one part. The DGX A100 can do all three within one system. Nvidia has shrunk the compute footprint just as Pure Storage has shrunk the storage side.”

The data requirements for those three capabilities can also be different, Fowler said.

“On the storage side, even if all the AI compute is consolidated on the DGX, you still need to be able to scale to handle not just the petabytes of data, but the hundreds of millions of transactions,” she said. “With our new FlashBlade//S:, we have doubled the density, performance and power efficiency of our storage. We’ve increased capacity to 2 petabytes but did not sacrifice the performance.”

The AIRI//S provides the muscle needed for AI workloads, Paikeday said.

“When we talk with customers and look at what they are trying to do with AI, many treat it as any other legacy workload,” he said. “But they often find very few of their models get deployed into production. A lot of the reason for that is the lack of infrastructure tools and applications. Customers tell Pure Storage and Nvidia they need both to develop AI. That is why they invest in AIRI.”

St. Louis-based solution provider and early Pure Storage channel partner World Wide Technology has done very well with Nvidia’s DGX technology and Pure Storage’s FlashBlade series arrays, said Jeffrey Fonke, primary storage practice manager for the solution provider.

“We look forward to building growth with AIRI,” Fonke told CRN. “We haven’t sold a ton of AIRI yet. But we’re looking forward to the advancements in the FlashBlade//S: arrays and future versions of the DGX technology to ensure the alliance between Pure Storage and Nvidia will move forward.”

Customers in such areas as life sciences, genomics, manufacturing and pharmaceuticals are purchasing solutions from World Wide Technology, Fonke said.

“We’re looking at ways to bring in new capabilities for these customers who are investing heavily in analytics and AI,” he said.

Justin Field, technical solutions architect for Pure Storage at World Wide Technology, told CRN that Pure Storage’s AIRI has benefited from Nvidia’s continuous advancement of its DGX platform.

“Customers are showing increased interest in solutions for the AI and machine learning space,” Field said. “And they want something that is easy to consume as a package with compute, storage and network like AIRI.”


This article originally appeared at crn.com

Got a news tip for our journalists? Share it with us anonymously here.
Copyright © 2018 The Channel Company, LLC. All rights reserved.

Most Read Articles

Log In

  |  Forgot your password?