Naveen Rao: 4D Computing, AI's Energy Wall & Beating Biology
Summary
Naveen Rao, co-founder of Unconventional AI, presents a new "dynamical computer" that uses physical oscillators to achieve 1,000x greater power efficiency than traditional GPUs. By leveraging dynamical systems theory and sparsity, this architecture eliminates the energy-intensive movement of data between memory and compute. Rao shares the first public results of their physical prototype, which generates images using mere nanojoules, aiming to solve the global energy crisis facing AI data centers.
Topics discussed
Introduction and Naveen Rao's background
Career history: Nirvana, Intel, and Databricks
Unconventional AI mission and company structure
The energy crisis in AI data centers
Biological inspiration: Brain efficiency vs GPUs
The problem of data movement and Moore's Law
Rethinking computation: Abstractions and physics
Dynamical systems and the metronome analogy
UNO model: Image generation via oscillators
Sparsity: Improving scalability and performance
First physical dynamical computer prototype results
4D Computing: A new architectural paradigm
Future implications: Efficiency, robots, and markets
Q&A: Path to product and ecosystem needs
Q&A: Porting existing models and architectures
Q&A: Team composition and software stack
Listen ad-free on Castria