Super Data Science: ML & AI Podcast with Jon Krohn · Jon Krohn

973: AI Systems Performance Engineering, with Chris Fregly

·1 hr 12 min·5 clips

He spent $6000 at Starbucks writing the 1000-page book that NVIDIA own documentation could not provide.

A decade-long booking finally lands. Jon Krohn brings on Chris Fregley and quickly makes the book the center of the episode: big, practical, and built for questions that do not fit into tidy AI hype. Fregley's background at AWS, Databricks, and Netflix gives the conversation a serious engineering weight without turning it into a resume tour. The book does most of the pulling here. Its thousand pages come across as a working reference for people who need to understand AI systems below the shiny surface. The cold open sets the stakes cleanly. NVIDIA's own documentation did not give Fregley the full performance picture he wanted, so he wrote the thing he needed. Then the size of the thing becomes the problem. Krohn points at the 175-item optimization checklist and, wisely, does not pretend a single hour can unpack it all. Fregley's answer is a workflow. When he analyzes a system, he puts the PDF into tools such as Cursor and Cloud Code. The move is simple: if the model may not know the book's detailed guidance, hand it the source material. Bigger context helps, but only if it leads to better questions and checks. The tools still have to meet the machine. Fregley brings up NVIDIA Insight and NVIDIA Compute as part of the setup that turns those checks into something useful. Krohn keeps pulling the conversation back toward listeners, asking which optimizations pay attention back fastest instead of walking through the whole checklist. The episode works best as a guided technical conversation for builders who want inspection habits, profiler evidence, and mechanisms. It ends with the show's usual thanks to sponsors and listeners, plus requests for sharing, reviews, subscriptions, and continued attention.

As heard by us

A dense, practical interview on AI systems performance engineering and how to turn it into a usable optimization strategy.

Chris Fregly and Jon Krohn keep the focus on AI systems performance engineering, not on model hype. The episode centers on what makes systems fast, measurable, and worth the cost, with Fregly drawing on experience from AWS, Databricks, and Netflix.

Read the full review in PlayNext →

Why you'd press play

A systems-level reset for how you think about AI performance engineering.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Super Data Science: ML & AI Podcast with Jon Krohn