Friday, April 19, 2024

Cerebras CS-1: a 400,000-core computer that replaces thousands of GPUs for AI

Cerebras System announced the launch of its first supercomputer, named “Cerebras CS-1,” propelled with its own Silicon Wafer Scale Engine (WSE) chip, dedicated to tasks related to Artificial Intelligence (AI). According to the company, the Cerebras CS-1 is the ‘world’s fastest deep learning computer system.’

The processor called Wafer Scale Engine, or WSE – announced back in August – has 1.2 trillion transistors, 400,000 compute cores, and 18 Gigabytes of high-performance on-chip memory. The dimensions of this chip are 220 × 220 mm; the area is 46,225 mm². It is 56.7 times larger and contains 78 times more compute cores than the largest GPU. More cores and more local memory allows a fast and flexible calculation, with less latency and less energy.

Cerebras CS-1
Cerebras calls it the world’s fastest deep learning compute system.

Following the details of the CS-1 Brain, it has 18 gigabytes of on-chip memory, nine petabytes of bandwidth, 12 Gigabit Ethernet connections to move data in and out of the system. It is powered with standard IEC C20 16A power inlets. Besides, as expected, the equipment includes a high-performance water cooling system to keep the components running at low temperatures, improving the performance and durability of this supercomputer.

The equipment is delivered assembled and ready to install and work. “The CS-1 is the industry’s fastest AI computer, and because it is easy to install, quick to bring up and integrates with existing AI models in TensorFlow and PyTorch, it delivers value the day it is deployed,” said Andrew Feldman, Founder and Chief Executive Officer, Cerebras Systems. “Depending on workload, the CS-1 delivers hundreds or thousands of times the performance of legacy alternatives at one-tenth the power draw and one-tenth the space per unit compute.

It is 26-inches (15 rack units) tall and fits in a third of a standard datacenter rack.
It is 26-inches (15 rack units) tall and fits in a third of a standard data center rack.

As for the software, Cerebras claims to have created its own platform that would allow developers to use their current machine learning libraries, and integrate them into their artificial intelligence workflows for use from CS-1.

Despite its size, WSE fits inside the CS-1, which is just 26-inches (66 cm) tall. That is, this computer occupies only one-third of the space of a standard rack of a data center, but would be able to replace hundreds or thousands of GPUs that would normally require dozens of racks.

Cerebras CS-1 is currently deployed at the Argonne National Laboratory, where it has been used to accelerate the studies that are currently being carried out there. Some of the topics they study there are how to better understand everything from cancer drug interactions to the properties of black holes.

Cerebras Systems is not the only company that creates computers or equipment dedicated to Artificial Intelligence since we can easily find other examples such as NVIDIA in addition to Amazon, which has also been working for some time to create its own processor oriented to this type of task.