Cerebras is a startup from California that chose a different path than most other manufacturers have done. In contrast to Nvidia, which every year brags about how they squeeze more and more transistors into smaller and smaller chips, they have chosen that bigger is better.
Nvidia’s largest gpu, the A100, measures a whopping 826 square millimeters and contains 54.2 billion transistors. Cerebra’s chip, named the Wafer-Scale Engine (WSE), occupies 46,225 square millimeters with a whopping 1.2 trillion transistors manufactured in 16-nanometer technology by the manufacturer TSMC. Now the company has delivered a finished result with its system CS-1, which had to compete against the Joule supercomputer in simulating flow mechanics. It was then specifically about combustion in a power plant.
CS-1 proved to be 200 times faster than Joule, which is ranked 81st on the list of the world’s fastest supercomputers. Cerebra’s system proved to be faster than reality. This means that if the simulation starts at the same time as a real reaction i begins in the power plant, the simulation is finished before the physical laws have time to carry out the process. Which makes it an incredibly powerful tool.
The giant circuit is cheaper than a supercomputer
In comparison to Joule, which costs hundreds of millions of kroner to build and takes up several server cabinets while consuming 450 kilowatts, CS-1 takes up a third of a server cabinet and consumes 20 kilowatts. At the same time that the cost of the circuit is only about tens of millions of kroner. Something that makes this a fantastic breakthrough that shows that there is much more to discover in technology than you might have thought. After all, it is important not to miss that an improvement such as CS-1 has proven to be, after all, means that we can use less resources but at the same time get an increased capacity. Which also means that costs are reduced and at the same time that less resources are used to run them. After all, supercomputers are something that increases our understanding in many different subjects and makes it possible to predict things before they happen as they can carry out incredibly advanced simulations.