Storage Unpacked 259 – Sustainable Storage in the World of AI with Shawn Rosemarin (Sponsored)

Storage Unpacked 259 – Sustainable Storage in the World of AI with Shawn Rosemarin (Sponsored)

Chris EvansAll-Flash, Guest Speakers, Pure Storage, Pure//Accelerate, Sponsored, Storage Unpacked Podcast

In this episode, Chris discusses the topic of building sustainable storage solutions with Shawn Rosemarin, Global VP of Customer Engineering at Pure Storage. AI and specifically Generative AI (GenAI) has become a hot topic over the past 12 months. Businesses are looking at projects to use AI internally for productivity gains, but also to drive additional business.

However, AI is still relatively expensive and requires huge volumes of training data. Training is an ongoing process that must react to changes in the data landscape, such as rights and permissions, and government regulation. With AI hardware being so expensive, it’s important to get the storage piece right, and that means having a scalable and cost effective solution. Shawn details how Pure Storage has focused on two aspects. First, the hardware, where DFMs (direct flash modules) have reached 75TB, with commitments to deliver 150TB and 300TB drives in the next few years. Second, the software management capability delivered through Purity, the operating system of Pure Storage hardware.

It’s clear that building cost and power-efficient flash devices will be a challenge for the wider industry, where the focus lies with consumer devices. Pure Storage believes it is well positioned to help customers and potentially hyper-scalers in their goals to deliver efficient storage for AI.

As Shawn highlights, this topic and more will be discussed at Pure Accelerate, to be held in Las Vegas from 18-21 June 2024. Check out the website where you can learn more.

Elapsed Time: 00:52:08

Timeline

  • 00:00:00 – Intros
  • 00:01:44 – We’ve been quiet on the topic of AI
  • 00:03:10 – AI has become cost-effective (sort of)
  • 00:04:00 – Efficient AI is a 10-15 year journey
  • 00:05:22 – AI technology needs to be efficient due to the resource demands
  • 00:06:41 – Data growth is currently growing at 30% per annum
  • 00:07:31 – Early mover may not be the best move with AI
  • 00:08:16 – 149 foundational models were released in 2023
  • 00:09:10 – Businesses will want to merge public and private data
  • 00:10:40 – Results accuracy is super-important
  • 00:13:30 – Trusted AI will be adopted in areas like security & vehicle evasive manoeuvres
  • 00:15:10 – Where will AI models be developed?
  • 00:16:37 – Model retraining will be required due to changing data ownership & permissions
  • 00:18:30 – Model training also needs to be resource efficient
  • 00:19:49 – $100 million to do the basic training of an AI model
  • 00:22:26 – How do you feed GPUs with adequate data to run at 100%
  • 00:24:10 – Edge devices could be used for AI processing
  • 00:25:18 – How will data centres need to evolve for AI?
  • 00:28:08 – Sustainability, regulation and jobs will all be issues in AI deployment
  • 00:31:05 – With HPC, many users built bespoke systems and that’s a problem for AI
  • 00:33:45 – How will businesses “industrialise” their AI projects?
  • 00:36:46 – Storage density will help resolve the operational issues of AI storage
  • 00:38:19 – SSD vendors’ main market is 2TB consumer SSDs
  • 00:39:32 – 300TB drives are great, but how will software manage the hardware?
  • 00:41:51 – Pure Storage DFMs will grow exponentially in capacity
  • 00:42:43 – Hardware engineering is cool again!
  • 00:45:15 – How will the hyper-scalers deal with massive storage growth?
  • 00:51:30 – Wrap Up

Related Podcasts & Blogs


Copyright (c) 2016-2024 Unpacked Network. No reproduction or re-use without permission. Podcast episode #xs2w