Let's Talk Shop
Let's Talk Shop
Storage Architect to AI: Is Your Data Performance Fast Enough?
Is storage is the new bottleneck in the age of AI? Elias Khnaser and Asad Khan, Senior Director of Google Cloud Storage, discuss this topic in depth. While all the spotlight is on fast, expensive GPUs and TPUs, Elias and Asad are back to basics.
In the past, CPU was never the bottleneck; slow storage was. Today, AI training and inferencing workloads require feeding high-cost GPUs/TPUs with data at an unprecedented speed to prevent them from sitting idle and wasting millions of dollars.
Key Takeaways:
► The shift: Why high-performance storage is now mission-critical for maximizing your ROI on massive GPU clusters.
► How Google Cloud is solving the data performance problem by moving beyond HDDs to intelligent SSD tiering.
► Deep dive into Google Cloud Storage solutions for AI, including Anywhere Cache and Rapid Store, designed to automatically handle caching, prefetching, and high-performance throughput across all zones without the customer having to worry about colocation.
► The importance of data APIs for researchers: object storage (GCS) vs. full POSIX compliance (Lustre).
► The truth: The best AI performance isn't just about the fastest chip—it's the correct configuration of GPUs, storage, and networking.
00:00:00 Intro & Guest Welcome: Asad Khan, Google Cloud Storage
00:01:19 GCS, Lustre, & the Full Google Cloud Storage Portfolio
00:02:00 Is Storage Dead? The GPU vs. Storage Conversation
00:03:12 The New AI Bottleneck: Why GPUs Sit Idle (Wasting Money)
00:06:39 From Cheap Scale to High-Performance Cloud Storage
00:08:22 The Two Dimensions of AI Storage: SSDs & APIs
00:10:37 Anywhere Cache: Automatic High-Performance Caching
00:13:15 How Storage Differs for AI Training vs. Inferencing
00:15:35 Rapid Store and Full POSIX Compliance with Lustre
00:18:26 The True Formula for AI Performance (It's Not Just the GPU)
00:20:39 Sony Honda Mobility Case Study: Lustre in Action
00:23:41 Traditional vs. AI Customers: Different Storage Priorities
00:27:07 The Future: Unlocking Insights from Unstructured Enterprise Data
00:33:40 Final Thoughts & Key Takeaways
Sign Up Now for my online course "The Cloud Strategy Master Class":
► On my web site: https://lnkd.in/gcxcrX and use promo code LINKEDIN20 to receive a 20% discount.
► On Udemy: https://www.udemy.com/course/cloud-strategy-master-class/
PODCASTS: Listen wherever you get your podcasts:
► Let's Talk Shop: http://letstalkshop.buzzsprout.com
► Reality Distortion Fields RDFs Podcast: https://youtu.be/88z1UiVaV00
Follow me:
► TikTok: @ekhnaser
► Instagram: @ekhnaser
► Twitter: @ekhnaser
► LinkedIn: https://www.linkedin.com/in/eliaskhnaser/
► Website: www.eliaskhnaser.com