In the ever-evolving landscape of technology and national security, partnerships between research institutions and tech companies play a pivotal role in advancing capabilities and setting new benchmarks. A prime example of such collaboration is the recent initiative between Sandia National Laboratories and Cerebras Systems. This partnership, now marking its second anniversary, has introduced a cutting-edge cluster composed of four Cerebras CS-3 systems. This cluster is poised to serve as a testbed for Sandia, significantly expanding research into artificial intelligence (AI) applications related to national security missions.
The Kingfisher cluster, as it is named, represents the first phase of an ambitious plan that envisions a total of eight Cerebras CS-3 nodes. This deployment is strategically aligned with the National Nuclear Security Administration’s (NNSA) Advanced Simulation and Computing (ASC) Artificial Intelligence for Nuclear Deterrence strategy. This initiative underscores the critical role of AI in national security, focusing on developing large-scale, trusted AI models using secure data from Sandia, Lawrence Livermore, and Los Alamos Laboratories.
### Advancing AI with the Cerebras CS-3 System
Cerebras Systems has brought to the table its third-generation wafer-scale engine architecture, known as the WSE-3. This technology is integral to the Kingfisher system and provides Sandia with unprecedented capabilities to explore AI’s potential. The unique architecture allows for expansive research into future AI applications, enhancing the existing ASC mission. Furthermore, it offers an opportunity to test how this architecture can be applied to Sandia’s traditional modeling and simulation workloads.
The Cerebras CS-3’s design addresses many of the challenges faced by traditional GPU systems, particularly those related to memory and power. According to Justin Newcomer, senior manager of the ASC program at Sandia, the CS-3 system is a game-changer for developing large-scale AI models. It aligns seamlessly with Sandia’s Advanced Architecture Prototype System program, known as Vanguard, which aims to push the boundaries of AI systems through strategic partnerships.
### The Innovative WSE-3 Architecture
At the heart of this collaboration is the Cerebras Wafer Scale Engine, or WSE-3. This novel architecture diverges from traditional semiconductor manufacturing processes. Typically, silicon wafers are divided into smaller individual processor dies after fabrication. However, in the Cerebras approach, the wafer remains intact, resulting in a single, large-scale processor that contains a staggering 900,000 processors optimized for AI and high-performance computing (HPC).
This integration allows for extremely high-performance computing within a single chip, with even greater potential when used in a cluster. The close proximity of processors to high-performance on-wafer SRAM memory facilitates rapid communication and enhanced memory bandwidth, offering a significant advantage over conventional AI accelerators.
### Generative AI and National Security
Sandia is actively engaged in several Generative AI projects aimed at enhancing capabilities for science and engineering applications in the national security domain. Siva Rajamanickam, Principal Investigator of the new BANYAN Institute, is enthusiastic about the prospects that the new system offers. The Kingfisher cluster will enable the evaluation of training and fine-tuning large multimodal models, critical for Sandia’s mission. The emphasis will be on exploring model accuracy, scalability, productivity, and power consumption during training workloads.
### DOE’s Commitment to AI Innovation
This initiative aligns with the Department of Energy’s (DOE) ongoing efforts under the Frontiers in Artificial Intelligence for Science, Security, and Technology (FASST) initiative. By integrating state-of-the-art architecture like the Cerebras CS-3 system, Sandia not only enhances its current capabilities but also lays the groundwork for pioneering advancements in AI. These advancements are crucial for supporting the DOE’s broader strategic objectives.
Jen Gaudioso, director of the ASC Program at Sandia, emphasized that the deployment of the Cerebras CS-3 system is a significant milestone in Sandia’s journey to lead in AI and machine learning innovation. The advanced testbed aligns perfectly with the DOE’s FASST initiative, enabling exploration and development of cutting-edge AI technologies vital for future national security missions.
### An Ongoing Commitment to Innovation
While the potential of AI to impact the NNSA mission is immense, traditional modeling and simulation remain critical components. The Kingfisher cluster, though primarily designed for AI, will also be used to explore traditional modeling and simulation workloads. James H. Laros III, distinguished member of technical staff and lead of the Advanced Memory Technology (AMT) program at Sandia, highlighted the importance of exploring the feasibility of using future versions of the Cerebras Wafer Scale Engine architecture for a combination of Mod-Sim and AI workloads.
Installed in October 2024, the Kingfisher cluster is already paving the way for exploration and innovation, made possible through the vital collaborations between national laboratories and industry. Gaudioso reiterated Sandia’s commitment to pushing the boundaries of AI research and development through this partnership with Cerebras Systems.
### Understanding the Role of NNSA
The NNSA, established by Congress in 2000, is a semi-autonomous agency within the U.S. Department of Energy. Its mission is to enhance national security through the military application of nuclear science. The NNSA maintains and enhances the safety, security, and effectiveness of the U.S. nuclear weapons stockpile, works to reduce the global danger from weapons of mass destruction, provides safe and militarily effective nuclear propulsion for the U.S. Navy, and responds to nuclear and radiological emergencies domestically and internationally.
The collaboration between Sandia and Cerebras Systems is a testament to the ongoing efforts to enhance national security through innovative AI research and development. As Sandia continues to explore the boundaries of AI capabilities, the Kingfisher cluster stands as a symbol of the potential for groundbreaking advancements that can shape the future of national security.
For more Information, Refer to this article.