Even one of the world's most powerful supercomputers, which takes 40 minutes to reproduce a single second of brain activity, is no match for the simple human brain. Researchers from Germany and Japan utilized K, the world's fourth-most powerful supercomputer, to mimic brain function. It took just under five seconds to complete one calculation, but the team estimates that it would take several months to build a computer as fast as your brain.
The human brain contains about 100 billion neurons connected by approximately 10^14 synapses. A synaptic connection is where two neurons connect with each other; one neuron sends an electrical signal through its body to the other neuron, which in turn sends this signal back out through its body. Neurons communicate with each other using chemicals called neurotransmitters. When one neuron needs to send a message to another, it produces these neurotransmitters. These neurotransmitters then travel across the synapse between the two neurons, causing them to switch on or off like switches on a circuit board. This allows the brain to control muscles, think thoughts, and experience feelings.
Neuroscience is a very new field of study, so much remains unknown about how brains work. However, scientists have been able to map the brain activity of people while they are awake and when they are asleep. They can also measure the blood flow to different parts of the brain during cognitive tasks to understand more about how it functions.
According to this trend, supercomputers will be able to do a million trillion computations per second (1 exaFLOPS) by the end of 2019, allowing for emulation of the human brain. Figure 1. Performance and power trends in supercomputers. Source: Cray
A petascale computer is one that processes 100 petabits (1015 bytes) of data per second. Computers currently range in performance from less than 1 kilobyte per second up to about 50 gigabytes per second. A computer that can perform a billion operations per second is known as a megacomputer, and a computer that can perform 10^12 operations per second is called a supercomputer.
When scientists use computers to simulate how the brain works, they need a lot of computation power. The human brain is estimated to contain about 10^9 neurons, with each neuron connecting to about 10^4 other cells. This means that the brain is capable of performing around 10^18 operations per second! Since modern computers are limited to about 10^6 operations per second, we need supercomputers to work on emulating the brain's behavior.
In 2001, IBM's Sequoia machine was the first to reach the petascale benchmark, processing 1 quadrillion calculations or "flops" per second. It used 80,000 processors and required 12 million watts of power.
The GBA supercomputer is valued 1.4 million dollars. A single second of human brain activity requires 1.4 million GB of RAM to compute. The computer has 64 cores with a total capacity of 128 gigabytes of memory.
Supercomputers are computers that are specifically designed for performing large-scale scientific research. They are usually built to handle many thousands of tasks at the same time. These tasks may include analyzing data from galaxy surveys, modeling nuclear reactions, or searching for extraterrestrial intelligence (SETI).
In 2013, Japan's National Institute of Informatics built the first exaflop-capable machine, the K computer. It has a peak performance of 100 petaFLOPS (100 billion floating point operations per second). It uses liquid cooling and consumes about 6 megawatts of power.
The K was built at a cost of $180 million and it takes 16 people full time to operate it. It was developed by Fujitsu in collaboration with IBM, HP, and Lenovo.
In 2014, China launched the Sunway TaihuLight which is considered by some to be the world's fastest supercomputer. It can perform up to 125 quadrillion calculations per second and its peak performance is estimated to be around 500 teraFLOPs.