Pure is announcing both the doubling of FlashBlade to 150 blades, an expansion of AIRI with the AI Data Hub, aimed at end-to-end use cases beyond simply training AI, and AIRI-as-a-Service, which will be delivered by Pure partner Core Scientific.
AUSTIN — At their Accelerate customer event here, Pure Storage announced enhancements to two of their core AI-related solutions, their FlashBlade file and object-based array for unstructured data and their AIRI [AI-Ready Infrastructure] offering with NVIDIA. They also announced an interesting partnership with Core Scientific around AIRI-as-a-Service.
FlashBlade has been expanded in large part to meet the needs of customers running AI applications.
“We have doubled the size of FlashBlade to 150 blades, scaling it to meet the needs of AI,” said Amy Fowler, VP of Strategy and Solutions for FlashBlade at Pure Storage, during the opening keynote on Tuesday morning. “That’s now nearly 8 PB in a single namespace.”
Customers have been asking Pure to increase FlashBlade’s size.
“If you look at the NAS market compared to SAN, there is much less flash penetration,” said Matt Kixmoeller, Pure Storage’s VP of Strategy. “We were the first flash NAS device out there, but we have been finding that there was a huge need at the top of the market for all-flash because of demands related to AI. We have been the leader in driving joint solutions around AI. We needed to make this bigger because many customers are up against the ceiling now.”
The expanded FlashBlade is now in Directed Availability.
“It is not quite complete yet,” Kixmoeller said. “We want to qualify use cases before customers buy.”
What is likely to announce customers and partners even more is the announcement of a major use case of FlashBlade, with the enhancement of their AIRI Data Hub architecture. Pure introduced the Data Hub last September. Built on FlashBlade, it was designed to power AI, analytics, and cloud-native applications by eliminating data siloes, and was the product of Pure’s longstanding partnership with NVIDIA and their DGX-1 GPU-based platform. The new AI Data Hub announcement is a logical extension of this.
“Data Hub is language we use to describe a consolidation of flash blades, and advanced analytics is one of the most aggressive places it is used,” Kixmoeller said. “The new AI Data Hub is a solution and reference architecture integrated with Kubernetes that builds on eight months of the AIRI Data Hub. It is an extension of AIRI to go beyond the training part of the pipeline where it has been focused – now essentially creating an end-to-end AI pipeline.
“The 150 blade FlashBlade is not exclusively for this, but it is a major use case,” he added.
“The AI Data Hub is such a valuable solution for customers in terms of providing an end-to-end view of the data workflow, which allows effortless end-to-end movement of very large data sets, especially from legacy data siloed environments so many of these organizations deal with,” said Tony Paikeday, Director of Product Marketing, AI and Deep Learning, at NVIDIA. “Models that used to take months to train just a few months ago can now be trained in under an hour.”
Paikeday said that this kind of capability is becoming a necessity in enterprises looking to make effective use of their data.
“This will empower customers to build better business relationships,” he said. “Every enterprise will need to become a supercomputing enterprise based on these capabilities that we are bringing to market.”
Fowler also announced AIRI-as-a-Service, which will be delivered by Core Scientific, a Pure partner that has developed an advanced AI and blockchain infrastructure-as-a-service using AIRI. In July, Core Scientific announced that it was partnering with Pure Storage to build a hosted AIRI to power the Cloud for Data Scientists
“We will actually deliver the service,” said Tony Chidiac, Core Scientific’s VP of Sales. “We own all the hardware, and in addition to Pure, are also partnered with Cisco and NVIDIA.
“This is all about saturating GPUs, something that is simply missing in the public cloud,” Chidiac noted. “What we get from FlashBlade provides 50 per cent savings compared to the hyperscalers.”
AIRI-as-a-Service is targeted at three main groups,
“The first is organizations that are just getting into GPU deep learning,” Chidiac said. “It makes no sense for them to spend a million dollars on AIRI. They can get 2 GPUs and FlashBlade from us on an Opex model for approximately $4000 a month.
The second bucket is larger, Fortune 1000-type companies whose AWS and Azure costs have ballooned.
“We are seeing a lot of this,” Chidiac indicated. “The efficiency of GPUs isn’t there in the public cloud.”
The third group is customers who could afford to buy AIRI, and have the space to host it, but are considering moving to the cloud.
“If they are looking to the cloud in order to scale, they sometimes ask whether it makes sense for them to just start with on off-prem service for this,” Chidiac noted.
Core Scientific has about 10 joint partners with Pure Storage.
“We and our partners can provide a deep expertise around accelerators,” Chidiac said. “One of our partners can accelerate Spark by 100x with no code rewrite!”