The human brain operates on roughly 20 watts despite containing about 86 billion neurons and trillions of synaptic connections. That makes it an extraordinarily energy-efficient biological computing system.
Researchers estimate that simulating a human-scale cortical model with around 20 billion neurons on an exascale supercomputer could require about 0.5 gigawatts (500 million watts) of electricity. That is roughly 25 million times the brain’s power consumption and enough electricity to power hundreds of thousands of homes.
The huge gap comes partly from how the brain processes information: neurons communicate through sparse, event-driven activity while memory and computation occur within the same interconnected biological network. Conventional computers, meanwhile, consume substantial energy moving data between separate processors and memory.
Scientists are developing neuromorphic computers inspired by the brain to narrow this efficiency gap. Technologies such as IBM’s TrueNorth and Intel’s Loihi use artificial neurons and event-driven processing, potentially paving the way for powerful AI systems that require dramatically less energy.
Source:
National Institutes of Health – NIH
Chetan Singh Thakur – Large-Scale Neuromorphic Spiking Array Processors – PMC