Pure Storage unveils AIRI Mini as ‘game changer’ to democratize AI in the enterprise

AIRI Mini is a joint engineering initiative with that provides an -focused solution with much less compute capability than AIRI, announced in March. It thus has a much lower entry point, and the companies think it will lead to strong interactions between infrastructure teams and data scientists.

SAN FRANCISCO – Among the flurry of announcements at the Accelerate event here Wednesday was the AIRI Mini. This followup to AIRI [AI-Ready Infrastructure] introduced two months ago, is based on a partnership with NVIDIA. Pure thinks that the AIRI Mini can have a massive impact on the adoption of AI in the enterprise, and make its use ubiquitous.

“The fundamental premise of AIRI Mini is to bring NVIDIAs DGX-1 servers together with FlashBlade and 100 GB networking to democratize AI within the enterprise,” said Matt Burr GM of FlashBlade at Pure Storage. “AIRI Mini really is a game changer. It delivers world class infrastructure, and replaces 215 racks of legacy infrastructure in one box that fits under a desk.”

AIRI Mini is powered by one Pure Storage FlashBlade, Pure’s platform architected for modern and AI, and two NVIDIA DGX1 servers, which deliver two petaFLOPS of performance. These systems are interconnected with 100GbE switches, supporting GPUDirect RDMA for maximum distributed training performance. AIRI also leverages the NVIDIA GPU Cloud software stack for maximum GPU performance, and the Pure Storage AIRI Scaling Toolkit, optimized for multi node training with DGX-1 and FlashBlade.

“You just have to turn the system on.” Burr said. “Yes, it really is that easy.” The idea is to allow data scientists to get their AI initiatives underway in hours rather than weeks or months.

“Customers told us they wanted design simplicity, and we found that we were looking to solve the same problems as Pure,” said Tony Paikeday, Director of Product Marketing at NVIDIA. “There was a need to deliver a simpler, faster solution that customers can stand up in record time and set themselves up for scale – and out of that came AIRI. Customers realize they need this infrastructure to make top line decisions, and they need a solution like AIRI to get them there.”

Paikeday said that NVIDIA’s benchmark testing of this produced very powerful results.

“The pace at which we could invest data into the DGX-1 was remarkable, as within 5 per cent,” he said. “It was like there was no network in between us. Those kind of benefits will enable data scientists the world over to get incredible results.”

The AIRI Mini facilitates close collaboration between data scientists and infrastructure teams that was simply not there before, said Chadd Kenney, CTO, Americas, at Pure Storage.

“CIOs told us that their infrastructure team have no ability to talk to the data scientists, with the result being that the data scientists are very constrained in what they can do,” Kenney said. “Our entry point with AIRI Mini gives infrastructure teams the ability to have this interaction with data scientists. That bridge we are building between data scientists and infrastructure teams is really important here. They can start off with the Mini in a relationship that allows them to scale.”

Kenney noted that the difference between the AIRI and AIRI Mini was simply the amount of compute, which is sizable.

“It makes for a much easier entry point,” he said.

“AIRI does not rest just on innovations in storage,” Burr said. “For users, I’m excited to announce we are extending both the AIRI and AIRI Mini to .” Both are now available with Nexus9000 100Gb Ethernet switches.

In addition, Burr announced a new engineered for Data Warehouse.

“It delivers 2.5x the performance of the next best system, and is extensible to modern analytics and AI when you are ready,” Burr said. It is fully integrated and validated by both Pure and Cisco.”

Both AIRI and AIRI Mini are available now.

Related Posts Plugin for WordPress, Blogger...

Leave a Reply

Your email address will not be published. Required fields are marked *