Cerebras and Ranovus Secure $45M DARPA Contract for Advanced AI Computing System

Cerebras and Ranovus Secure Major DARPA Contract

Cerebras Systems, a Silicon Valley-based AI chip company, and Canadian chip startup Ranovus have been awarded a $45 million contract from the U.S. Defense Advanced Research Projects Agency (DARPA) 1

1

. The contract aims to develop a high-performance computing system that will significantly speed up connections between computing chips, potentially revolutionizing AI processing capabilities.

Innovative Wafer-Scale Technology

At the heart of this project is Cerebras' WSE-3 chip, a wafer-scale processor optimized for AI workloads. Unlike conventional chips that are slightly larger than a postage stamp, the WSE-3 is about the size of a dinner plate, measuring roughly a foot wide 2

2

. This massive chip boasts impressive specifications:

• Four trillion transistors
• 900,000 cores
• 44-gigabyte onboard SRAM memory

Cerebras claims that this design allows for more efficient AI processing compared to traditional graphics cards, as the large onboard memory can often store all data generated during AI model processing, reducing the need for external data transfers 2

2

.

Integration of Co-Packaged Optics

A key aspect of the DARPA project is the integration of Ranovus' "wafer scale co-packaged optics" technology with Cerebras' chips. This innovation aims to use light, rather than electrical signals, to transfer information between chips more quickly and with lower power consumption 1

1

. The co-packaged optics technology eliminates the need for standalone pluggable transceivers by integrating them directly into the processors, potentially increasing efficiency and network capacity in AI clusters 2

2

.

Project Goals and Implications

The collaboration between Cerebras and Ranovus under the DARPA contract has ambitious objectives:

1. Develop a computing system capable of simulating complex battlefields in real-time 1

   1

   .
2. Achieve processing speeds 150 times faster than current systems while reducing power consumption from 30 watts to just 3 watts 1

   1

   .
3. Deliver a platform for high-fidelity simulations of challenging physical environments and large-scale AI workloads 2

   2

   .

Broader Context and Future Prospects

This contract comes at a significant time for Cerebras, as the company has already filed for an initial public offering and is positioning itself to challenge Nvidia in the AI chip market 1

1

. The project also highlights the growing interest in optical computing solutions, with various startups seeking funding to tackle the challenge of integrating optical connections with computing chips 1

1

.

Cerebras is also involved in another DARPA project called the DRBE program, which aims to create a virtual environment for testing wireless communications systems 2

2

. This multi-faceted engagement with DARPA underscores the company's growing importance in advanced computing research and development.