Do you remember the Intel’s original Itanium processor, code-named McKinley? It was considered to be massive chip with 221 million transistors when it was first launched in 2002. Now in 2019, Cerebras, a computer system company that is dedicated to accelerating deep learning is all set to launch the largest chip ever at the Hot Chips Conference at Stanford University. The company claims that the chip is going to be as large as computer’s keyboard. As per the company, the chip requires 46,225 square millimeters of silicon and 1.2 trillion transistors which is 5,000 times the transistors that McKinley had. Hence it would be a rectangular chip that would measure 8 by 9 inches.
The Cerebras Chip is not designed for your PC, but the company uploaded the picture of the ship, comparing its size with the keyboard. For now, the company has not revealed the crucial details such as amount of memory of the chip or the manufacturing cost. But the chip is manufactured in a relatively old 16nm process technology from TSMC. The brand designed its chip for deep learning and characterized the chip as a “wafer scale” implementation of a neural network, essentially clustering the logic, interconnect and memory onto a single piece of silicon.
The former chief hardware architect for Sea Micro, later acquired by AMD, as well as the Data Centre Server Solutions business at AMD, Mr Sean Lie said that designing a single chip save manufacturing cost because, in a single chip you get all of it at vastly lower power and less space.
As the size of chip is large or for cooling, a cold plate is attached above the silicon, using vertically mounted water pipes to cool the chip directly.