British scientists have launched a supercomputer with 1 million cores that simulates the human brain
On November 2, researchers at Manchester University launched the most powerful supercomputer whose architecture simulates the human brain. It employs a million ARM9 cores that can handle 200 trillion operations per second. Computer designed for 20 years, and another 10 years collected. The construction work started in 2006, and the total amount spent on the project was £ 15,000,000.
How does it work
The main difficulty of such a project is to create iron, which can cope with a multitude of parallel processes in much the same way as a biological brain does. For example, the human brain consists of approximately 100 billion neurons. The number of connections between these neurons is astronomical. The researchers plan to create a model of one billion neurons using their supercomputer , which will be processed in real time. Despite the fact that this is only one percent of the human brain - a step for computer science is huge. So far, no supercomputer could cope with this task.
The main decisions in the supercomputer are related to the architecture of iron.. It is called SpiNNaker (Spiking Neural Network Architecture). It has a million cores divided into 57 thousand nodes. Each has 18 cores and 128 megabytes of SDRAM memory, which is located separately from the matrix with chips. There are several ways to transfer information between the cores, but the main one is in packets of 5–9 kilobytes.
All nodes are connected to an asynchronous infrastructure. Mobile processors and memory are chosen for energy reasons - in such a powerful system it is one of the main resources for which performance had to be sacrificed.
As the researchers say , the knot of 16 cores could cope with the task, but the number was increased for fault tolerance.
The principles of interaction between the processors relied on neurobiology. In conventional computers, the amount of data sent is much higher, because they contain information about how and where the system should deliver information.
In the SpiNNaker architecture, this was avoided. Data packets contain only information about the source, and the infrastructure sends them to the destination. As the Motherboard Edition explainssuch an infrastructure is like working mail without post offices - everyone has a personal postman who himself knows which address to deliver the parcel to.
Steve Furber, a computer engineering professor and project researcher, says: “SpiNNaker completely reinvents the way computers work. We managed to create a machine that works more like a brain than a computer, and this is just incredible. ”
What will be used for
Now the computer will be used for academic and scientific research. For example, scientists have already recreated the area of the cerebral cortex consisting of 80 thousand neurons, which are responsible for the processing of sensory information. This model helped the robot in real-time to process video information in order to navigate the difficult terrain.
Also, the supercomputer can simulate the basal ganglia - a part of the brain that suffers during the development of Parkinson's disease. It is likely that the model will help neuroscientists to advance in the fight against this ailment.
As Furber added in his statement, “with the help of SpiNNaker, scientists can uncover many secrets of the brain by exploring simulations of enormous scale. Robotics will be able to create the most powerful neural networks for mobile robots so that they can move, talk, work flexibly and less energy-intensively. ”