Cerebras Systems Introduces "The World's Fastest" CS-1 Supercomputer Optimized for Computing in AI and Machine Learning. It was first installed in the Argonne National Laboratory (USA).

At a U.S. Department of Energy research center, a 400,000-core system will model black hole collisions, help develop cancer drugs, and look for more effective treatments for brain injuries.

According to hitech.vesti.ru, the main feature of the CS-1 is the Wafer Scale Engine chip. Each WSE chip has 1.2 trillion transistors (the “ordinary” chip has no more than 10 billion), which are organized into 400,000 AI cores and 18 GB of internal SRAM. The memory bandwidth is 9 PB / s.

Each CS-1 component was designed from the ground up to solve problems associated with neural networks, said VentureBeat Cerebras Systems CEO Andrew Feldman. This provides higher performance with less energy consumption and the amount of occupied physical space.

The WSE chip is 56.7 times larger than the largest video accelerator, and contains 78 times more computing cores. According to Feldman, CS-1 will be able to cope with the task in the field of AI, which today takes months, in minutes. And the work that takes weeks will take only a few seconds.