This computer has been named Cerebras CS-1, and no, it will not be used for ‘Minecraft’ or ‘Fortnite’ in 4K at 60 fps, but for tasks related to artificial intelligence and its motherboard carries an impressive chip larger than an iPad in dimensions and with 400,000 cores, 18 GB of memory and 1.2 billion transistors.
According to information provided by the same company, Cerebras CS-1 is a super computer that they offer as a “complete solution” designed to be added to a data center to handle artificial intelligence workflows. According to the company, today there is no other system in the world that is faster in processing artificial intelligence workloads.
Andrew Feldman, founder and CEO of Cerebras Systems, explained that the CS-1 is delivered ready to install and work, and is compatible with existing artificial intelligence models such as TensorFlow and PyTorch.
The central part of the CS-1 is the Wafer Scale Engine (WSE), which is the first to break the billion transistor barrier, specifically 1.2 billion transistors. Its impressive size is 46,225 square mm, about 56 times larger than the largest GPU ever made. Despite its size, WSE fits inside the CS-1, which has a size of just 67 cm in height. That is, this computer takes up only a third of the space of a standard data center rack, but would be able to replace hundreds or thousands of GPUs that would normally require dozens of racks.
The CS-1 includes 400,000 processing cores, 18 gigabytes of on-chip memory, 9 petabytes per second of bandwidth, 12 Gigabit Ethernet connections to move data in and out of the system, and a power consumption of just 20 kilowatts. According to specifications, this whole monster includes a high-performance water cooling system to keep this large chip and platform running at low temperatures and prevent overheating.
As for the software, Cerebras claims to have created a proprietary platform that would allow developers to use their existing machine learning libraries, and integrate them into their artificial intelligence workflows for use from CS-1.