Naveen Rao: 4D Computing, AI's Energy Wall & Beating Biology
Naveen Rao, CEO of Unconventional AI, joins the All-In hosts to explain why current AI hardware wastes most of its energy moving data and how a new chip design could fix that.
The gist
-
AI energy use is already severe, with Google consuming 12 gigawatts monthly for AI, and Rao says power will be the hard limit within about 3 years.
-
The human brain runs on 20 watts and a squirrel's on 8 milliwatts, showing biology solves efficiency through physics rather than the binary logic modern chips use.
-
Rao's prototype oscillator-based chip uses 500 nanojoules per image versus millijoules on GPUs, a 1,000x or more efficiency improvement already demonstrated in an open-source model.
-
Rao says a full rack-based data center product is about 2 years away, though it requires retraining models rather than a simple drop-in swap.
Also mentioned
2:10
Nirvana Systems
· "I actually founded the first AI chip company called Nirvana Systems. This was in 2014. If anyone remembers..."
2:33
Intel
· "I sold the company way too early to Intel, but I started and ran the AI group at Intel."
2:59
ChatGPT
· "After ChatGPT happened in 2022, we were kind of the best game in town for people to start building their..."
3:23
Unconventional AI
· "I want to tell you about Unconventional AI, which is rethinking the foundations of how a computer works...."
◎
Related episodes
Episodes covering similar topics
More from All-In Podcast