Pure Storage expands machine learning range with AI Data Hub

Plus, its AIRI hardware platform is now available as a service

by Max Smolaks 18 September 2019

American all-flash storage specialist Pure Storage has introduced the AI Data Hub – an end-to-end pipeline for artificial intelligence workloads that brings together hardware and software from Pure and its partners, most notably GPU powerhouse Nvidia.

The Hub architecture can handle all stages of machine learning projects, from data cleansing and tagging through to modeling, training and production, all wrapped in a workflow based on Kubernetes.

Pure also announced that its AIRI appliance – essentially Nvidia DGX-1, a GPU-based supercomputer, married with high density FlashBlade storage boxes and 100GbE switches from Arista – is now available as a service from a number of data center partners, including Core Scientific.

The announcements were made at the Pure Accelerate conference in Austin, Texas.

Pure Storage was established in 2009 to transform consumer-grade flash memory into affordable, enterprise-ready solutions. Even though it sells hardware appliances, most of the company’s intellectual property is contained in its software and data reduction algorithms.

In recent years, Pure has been repositioning itself as a supplier of hardware platforms for AI workloads, launching AIRI (AI-Ready Infrastructure) in 2017.

Now, AIRI itself has been integrated into a larger set of products called the AI Data Hub. Pure says it offers an end-to-end AI pipeline solution that gives enterprise customers what they need to design, develop and deploy AI quickly, and at scale.

Besides AIRI, elements of the AI Data Hub include RAPIDS – open source software for accelerating data science and analytics on GPUs, originally developed by Nvidia – as well as FlashStack, a converged infrastructure product Pure developed in cooperation with Cisco, and T4 GPUs that are perfect for running AI algorithms in production.

“With AIRI powered by Nvidia DGX systems and Pure Storage FlashBlade, Nvidia and Pure have been making it faster, easier and more cost-effective for every enterprise to deploy and manage AI at scale,” said Charlie Boyle, vice president and general manager of DGX systems at Nvidia.

“The AI Data Hub increases the value of our joint solutions by accelerating data science workflows, powered by Nvidia CUDA-X AI software and DGX systems for faster iteration, with greater developer productivity and accelerated time-to-insight.

For organizations that don’t want to buy appliances of their own (rough calculations suggest a single AIRI costs upwards of $1 million), Pure now offers AIRI-as-a-Service, delivered through a network of infrastructure partners.

“Modern organizations want to leverage the cloud, but most public clouds available today aren’t optimized for the requirements of AI. We’ve designed a cloud purpose-built for data-intensive applications like AI and deep learning, built on top of AIRI,” said Kevin Turner, CEO at Core Scientific.