Meet the Cerebras CS-1 supercomputer, which is 200 times faster than any other and uses a gigantic chip

Its manufacturers and creators showed that the chip is 60 times larger than that of a conventional supercomputer. And much faster.

A supercomputer it is always a project that means a great advance for technology. And this one does meet the conditions to take steps forward in the scientific world. It is a project of researchers from the National Laboratory of Energy Technology (NETL), in conjunction with the company Cerebras. Both organizations built a machine that exceeds 200 times the speed of supercomputers fastest in the United States.

First of all let’s clarify that this computer is not for common use. This type of artifact is intended for scientific functions and macro science projects. It is identified as Cerebras CS-1. And it is capable of solving a large, sparse, and structured system of linear equations, Slash Gear reported.

This type of equations that can be quickly solved by this machine, is used to model physical phenomena, such as fluid dynamics. In more understandable terms for those of us who are not experts in this matter, the phenomena it shapes are used in weather forecasts, for example.

They would also serve to find the best shape and model for an airplane wing. In addition to predicting temperatures and finally it is able to predict radiation levels in a nuclear power plant. To work this way it uses a finite volume method on a regular three-dimensional mesh.

The structure of the supercomputer

The technology company Cerebras says that there are three key factors for the level of speed to develop in this device. First is the bandwidth. In addition to the low latency of the communication tissue on the wafer. And finally the processor architecture optimized for bandwidth computing.

The chip, the largest in the world, measures about 72 inches square. It is 60 times the size of a conventional chip. Its speed was specifically compared to the NETL Joule 2.0, which has a cluster of 84,000 CPU cores. And when taking the measurements it was 200 times faster.