Utilizing Tech - The Podcast Series about New and Emerging Technologies · Tech Field Day - Part of The Futurum Group

07x01: Proving the Performance of Solidigm SSDs at StorageReview

·36 min·3 clips

Jordan reveals how AI helps detect focus issues in astrophotography in real-time, preventing wasted weekends of data capture.

This episode of Utilizing Tech focuses on AI data infrastructure and features Jordan Ranos, the Advanced Workload Specialist from Storage Review. Host Stephen Foskett and co-host Ace Stryker from sponsor Solidigm explore real-world testing of storage technology beyond vendor specifications. Jordan details a specific project using Solidigm SSDs for an edge AI workload in astrophotography. The team deployed a ruggedized Dell XR7620 server with four Solidigm QLC SSDs to a remote location for capturing space imagery. They captured raw images at 62 megapixels each and combined this data with Hubble legacy data to train a convolutional neural network. This CNN performs real-time denoising and sharpening, providing immediate feedback on data quality issues like focus or lens dew. The project processed over 2000 neural network layers, requiring enough VRAM to span four NVIDIA H100 GPUs during the training phase. Jordan explains the system used a mix of Solidigm's 60TB and 7.68TB U.2 NVMe SSDs for high-capacity field storage. A key insight was the efficiency of physically shipping these high-capacity drives back to the lab versus uploading petabytes of data over limited bandwidth. He connects this astrophotography use case to broader applications like self-driving cars, oceanic exploration, and retail, where edge inferencing can process data locally and send only metadata back to a central data center. The conversation reveals that the SSDs operated successfully in a blizzard, beyond their official temperature ratings, highlighting their ruggedness. Jordan also discusses Storage Review's extreme testing methods, like calculating 202 trillion digits of Pi to hammer SSDs with petabytes of writes and validate endurance. He outlines the complex challenge of benchmarking storage for AI, noting different phases like data ingest, training, and checkpointing have unique performance requirements. The tone is conversational and educational, blending technical deep dives with practical anecdotes from field testing. The style is interview-based, with the hosts prompting the guest to share specific experiences and insights. Listeners interested in edge computing, AI infrastructure, or real-world tech applications will find this episode compelling. Those seeking only high-level vendor marketing or theoretical discussions might find the detailed technical benchmarking less engaging.

As heard by us

A practical look at AI data infrastructure and edge deployment, grounded in real-world testing.

This episode keeps its focus on AI data infrastructure and the practical questions that show up once hardware leaves the lab. Jordan Ranos of Storage Review brings a grounded view, and the strongest example is the rollout to 1,500 retail locations, where inference stays at the…

Read the full review in PlayNext →

Why you'd press play

You want a reality check on AI deployment before the edge rollout.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Utilizing Tech - The Podcast Series about New and Emerging Technologies