Several international media echo the
progress made by Cerebras Systems, the Silicon Valley-based start-up
that is developing the world’s largest chip for intelligence work
The news came after Cerebras stated
two days which has the ability to interlist 192 chips to form extensive
neural networks, without affecting the efficiency and computational potential of the
performance of AI functions.
In a way, the Cerebras microprocessor can
to double the processing capacity of their chips, unlike
current computer systems that need twice the energy to multiply
their level of performance by two.
Supported by its advanced developments in
semiconductors specially designed for AI processes, Cerebras
face industry leaders such as Nvidia Corp and Alphabet Inc. Google obtained
$475 million of venture capital and signed contracts with major
pharmaceutical companies to use their chips to speed up the processes of
R&D to get new drugs.
Innovations that can change the industry
In the common microprocessor manufacturing processes,
hundreds or even thousands of chips are manufactured on 12 silicon discs
inches called wafers that, in the next phase, are divided into chips
Individual. The manufacturing process of Cerebras, however, uses all the
wafer, so that the new chip has the ability to accumulate much more data by
at the same time.
Researchers working on new
AI developments have artificial intelligence models in neural networks
which are too extensive to be managed by a single chip, so they have
to divide tasks between multiple microprocessors.
The most extensive neural network that exists has only one
small fraction of the complexity of the potential of a human brain, and in addition,
uses much more energy because the systems that activate them weaken their
performance as more chips are embedded in this network.
In any case, Cerebras is not alone in
race for a chip-scale machine learning model. In 2020, the
Intel connected 768 of its Loihi chips to form a cluster of 100 million
neurons and rodent-like intelligence. In 2019, Loihi put to the test
its great potential by obtaining 100 billion synapses whose potential
comparable to that of a rat brain.