solutions
Sep 30, 2021

Our Expanding Universe of AI Infrastructure

Our Expanding Universe of AI Infrastructure

Posted by

Jeff Denworth

Big ideas require big compute power.

Here at VAST Data, we love the research that’s published by one of the world’s leading investment management groups, Cathie Wood’s ARK Invest. In ARK’s 2021 Big Ideas Report, the ARK team put a fine point around the significance of AI as ‘the most important software breakthrough of our time.” Here are some of the highlights of the report:

  • With deep learning, software is now essentially writing itself

  • While the cost of training reduces 37%/year, AI models are growing in size 10x annually

  • By 2025, it will cost approximately $100M to train an AI model

  • Data center spending on AI processors will increase by a factor of 4x by 2025

  • AI will add $30 trillion in value to the global equities market over the next 15-20 years

images

Source: Big Ideas 2021, ARK Invest

This cambrian explosion in data center infrastructure has resulted in a cambrian explosion of data. The VAST Data Platform has become the easy button for customers to build scalable, performant and affordable all-flash infrastructure to power next generation AI training.  But storage alone is not a solution, so we work diligently to qualify and optimize modern computing platforms to work seamlessly with our DASE architecture.

Today, we’re happy to announce that VAST has partnered with Graphcore to showcase an integrated solution that brings the power of Graphcore’s IPU and the simplicity and affordability of scalable all-flash that is the hallmark of VAST’s Data Platform. As our respective teams have found each other working with more and more common customers, it made perfect sense for us to combine efforts and take the guesswork out of building large-scale AI training infrastructure.

images

Released in 2020, Graphcore’s M2000 IPU is a powerful accelerator designed from the ground up for AI. The MPU2000 features 1 PetaFlop of compute and up to 526GB of memory in 1U. Graphcore clusters these systems into IPU-PODs to help customers build scalable infrastructure and deploy scalable deep learning codes to unleash new AI breakthroughs.

VAST powers IPU Pods with the game-changing innovations that has led to VAST’s DASE architecture becoming the platform for scalable computing in the intelligent enterprise.

  • Scale datasets and performance, linearly, to exabyte and exascale proportions

  • A next-gen approach to storage efficiency, making flash affordable for all your data

  • Accelerated NFS access that provides direct RDMA all the way to the processor

  • Simple scale-out NAS appliance that eliminates the need for complex HPC storage

For customers, Graphcore and VAST are better together. When smart infrastructure partners work together to simplify and accelerate the AI Infrastructure deployment experience, data science wins.

For more information on our work with the Graphcore team, reach out to us here.

More from this topic

Learn what VAST can do for you
Sign up for our newsletter and learn more about VAST or request a demo and see for yourself.

By proceeding you agree to the VAST Data Privacy Policy, and you consent to receive marketing communications. *Required field.