Intel has announced what for them is its first system capable of mimicking the human brain through neuromorphic computing . A total of 64 chips based on the company’s Loihi prototypes that together amount to the calculation made by 8 million human neurons. In its initial phase, this new system will only be accessible to Intel partner researchers.
To put the figures in context, it is estimated that the human brain contains about 80 billion neurons. Ten thousand times more than this Intel chipset can calculate. But that does not detract from it, they say it can calculate 1,000 times faster and 10,000 times more efficiently than chips specialized in specific topics such as scattered coding or graph search.
Processors with neuromorphic engineering
The attempt to imitate the human brain already gives some clues as to what the objective of this new Intel processor system is. Indeed, ‘Pohoiki Beach’ is designed for areas of artificial intelligence . It has been designed using the neuromorphic computing architecture, which seeks to mimic the biological neuronal system to obtain better calculations. In principle, it manages to adapt better to the changes and uses the knowledge it already has to learn exponentially over time.
Intel is not the only company that has meddled in this processor technology. Qualcomm has already produced neuromorphic chips . On the other hand, DeepMind, Alphabet’s machine learning company, uses deep learning methods that allow systems to learn on their own through the experience they acquire.
The strong point of Intel’s Loihi chip is energy consumption , as indicated. According to the first tests performed by some of Intel’s partners, this Loihi chip uses a hundred times less energy than a normal processor when performing machine learning calculations. When scaling the system they have only seen a 30% increase in energy consumption, compared to more than 500% that usually occurs using other systems.
The 8 million neurons are just the beginning, Intel promises. They hope to release a new version of ‘Pohoiki Beach’ called ‘Pohoiki Spring’ by the end of this year. With this new version, they want to scale the system to achieve a calculation equivalent to 100 million neurons , still far from the 80 billion human brain, but a little closer.