Researchers have devised an unprecedented AI supercomputer: it will simulate the human brain on a scale

Spread the love

The brain is one of the most complex and sophisticated organs in the human body. We have been studying it for a long time and we still know very little about it. However, after much scientific effort, we have managed to come up with some parallels, such as quantum computing and artificial intelligence general.

What if we could, through technology, achieve the performance of a human brain by imitating its biological processes? Researchers at Western Sydney University believe it is possible. And their theory has not only remained in the books, but they have launched a project to make it a reality.

Say ‘hello’ to the DeepSouth Supercomputer

The International Center for Neuromorphic Systems (ICNS) of the aforementioned institution explains that they are working on the first computer system capable of the networks of the human brain thanks to a neuromorphic system. It has been called DeepSouth Supercomputer and is expected to be operational in 2024.simulate at scale

Unlike conventional computer systems, DeepSouth has been designed from the ground up to function as if it were a huge neural network. It is a completely new platform that promises to establish a new paradigm for how computing loads are managed and, they say, will come with many tangible advantages.

First, the supercomputer architecture will be able to emulate the human brain to reach peaks of 228 trillion synaptic operations per second. The ICNS assures that this is equivalent to the estimated rate of operations that the human brain can perform. Plus, it will be tremendously efficient.

According to researchers, the human brain can process information equivalent to one exaFLOP per second with something like 20 watts of power. DeepSouth’s neuromorphic engineering promises to process an equivalent amount of data with much less energy. The benefit? Improve the development of artificial intelligence.

“The simulation of spiking neural networks on standard computers using graphics processing units (GPU) and central processing units (CPU ) multicore is too slow and consumes a lot of power. Our system will change that,” explains ICNS director André van Schaik. DeepSouth promises to be very different.

There are some important details that have not yet been released, such as what hardware equipment will be part of this system. Apparently, we won’t see a traditional architecture based on CPUs, like the Intel Xeon, and GPUs, like the NVIDIA H100, to run DeepSouth. In April 2024, if everything goes according to plan, these doubts will be resolved.