IBM Simulates 4.5 Percent of Human Brain Activity Using Blue Gene Supercomputer
In November 2011, IBM researchers led by Dharmendra Modha at IBM Research Almaden demonstrated a cortical simulation on the Blue Gene/P supercomputer that modelled approximately 4.5 percent of human-scale neural activity, using 147,456 processors to represent 1.617 billion neurons and 8.87 trillion synapses.
Background
Simulating brain-scale neural activity had been a goal in computational neuroscience for decades, but the numbers involved made it almost comically difficult. The human brain contains roughly 20 billion neurons, connected by something in the order of 200 trillion synapses. Even a mouse brain, orders of magnitude smaller, required hundreds of processors to model at any meaningful level of detail.
The field had made incremental progress through the 2000s by working up through species. Researchers found they could use the processor counts needed for each animal as a kind of ladder: 512 processors for a mouse-scale simulation, around 20,000 for a rat, around 24,500 for a cat. Each step revealed how the hardware requirements scaled, and each step made the distance to a human-scale model look larger. The honest conclusion from that work was that simulating the full human cortex would need somewhere close to 800,000 processors. No machine at the time came close to that.
What the field lacked was a direct experimental demonstration at primate scale. Extrapolating from rodents was one thing. Actually running a simulation that occupied a meaningful fraction of human neural complexity, on real hardware, with real results, was another.
What happened
In November 2011, Dharmendra Modha and his team at IBM Research Almaden ran a cortical simulation on IBM’s Blue Gene/P supercomputer that represented 1.617 billion neurons and 8.87 trillion synapses. That worked out to roughly 4.5 percent of human-scale neural activity. To do it, they used 147,456 processor cores running in parallel.
Blue Gene/P was one of the most powerful supercomputers available at the time, built around a massively parallel architecture designed for exactly this kind of work: problems where the data can be spread across thousands of processors rather than passed through a single fast one. Even so, running the simulation pushed it hard. The team presented the results at SC11, the Supercomputing conference held in Seattle in November 2011.
The simulation was not a model of thought or consciousness. It represented the physical structure of neural tissue, mapping neurons and the synaptic connections between them at a scale that had not been attempted before in a primate-range system. The point was not to recreate brain function but to find out what the hardware cost actually was when you stopped extrapolating and started measuring. Those measurements, across mouse, rat, cat and partial human scales, gave the research community a concrete set of numbers to work from, and a clearer sense of what it would take to reach full human-brain scale.
Why it mattered
The simulation established a concrete computational benchmark for whole-brain modelling, quantifying the processor counts required to scale from rodent to primate to human cortical networks and demonstrating that existing supercomputer architectures could approach, but not yet reach, full human-brain scale. It gave the neuromorphic computing community a measurable target and contributed directly to the rationale for IBM's SyNAPSE programme, which sought brain-inspired chip architectures as an alternative to von Neumann hardware for neural-scale workloads. By publishing scaling figures across species (mouse, rat, cat, and partial human), the team provided a reproducible methodology for comparing progress across future simulation efforts.
People
Organisations
IBM Research Almaden, IBM
Sources
- Graphic Science: IBM Simulates 4 Percent of Human Brain.Scientific American.Secondary
- Cognitive Computing group page, Dharmendra Modha, IBM Research.IBM Research.Official
Cite this page
AI Achievements. (2011). IBM Simulates 4.5 Percent of Human Brain Activity Using Blue Gene Supercomputer. Retrieved 2026-08-22, from https://achievements.ai/milestone/ibm-simulates-the-human-brain-4-5-percent
@misc{achievements_ibm_simulates_the_human_brain_4_5_percent,
title = {IBM Simulates 4.5 Percent of Human Brain Activity Using Blue Gene Supercomputer},
author = {{AI Achievements}},
year = {2011},
url = {https://achievements.ai/milestone/ibm-simulates-the-human-brain-4-5-percent}
}