Man and Machine

An average human brain contains about 100 Billion neurons, although, at this point in your life, you may have fewer.

An average neuron has 1500 signal connections to other neurons.There are about 0.15 quadrillion of these "synapses" in the brain.

The senses transduce external stimuli into neural action which modifies the states of connected neurons. Each of the 100 Billion neurons in your brain is in a changing state at every instant.

Your friend, the neuron

Japan's "Earth Simulator" supercomputer performs 38 Terra flOps / second.

1 Terra flop is 1 000 000 000 000 floating-point operations

This is about .0003% as much as would be required to completely model every neuron in the brain. See Brain Computing to understand how this estimate is achieved.

To Simulate 1 Brain requires 120,000,000 Pentium-4 Processors
(each P4 operating at 3Ghz)

When scientists focus on isolated tasks (like playing Chess or recognizing speech) and use algorithms which exploit specialized knowledge, computer scientists have been able to build machines which outperform Man.

Neural Networks are now successfully used to process information on highly-constrained problems, but a complete simulation of the brain remains elusive.

top

Brain Computing

It is interesting to estimate how many arithmetic operations might be required to simulate an average human brain. If this simulation is done in the simplest possible way - by adding the contribution of every synapse to every neuron we could proceed as follows:

Number of synapses in a brain =
synapses per neuron * neurons in a brain =
1500 * 100 000 000 000 = 1 500 000 000 000 000 = 1.5 x 10¹⁴

Number of calculations per synapse = 2 (read current state of synapse, add to sum for connected neuron. Note these are integer operations (not floating point))

Number of calculations per second per synapse = 1000 (allows for a maximum firing rate of 500Hz for each synapse, which is about 2 to 5 times higher than normally recorded)

Total number of integer operations per second = 1.5 x 10¹⁴ * 2000 = 3.0 x 10¹⁷

Brain:
300,000,000,000,000,000 iOps / second

"Earth Simulator" supercomputer performs 36 Terra flOps / second. This is about 15 000 times better than a P4. Assuming a floating point operation is performed at roughly the same speed as an integer operation (it is actually slower), we have...

Earth Simulators required to model 1 brain = 3.0 x 10¹⁷ / 3.6 x 10¹³ = 8333. This is roughly equivalent to 120 million pentium 4 computers!

1 Brain = 8333 State-of-the-art Supercomputers

top