Sandia National Laboratories launched Kingfisher AI system on huge Cerebras WSE-3 chips

The US Department of Energy (DOE) Sandia National Laboratories (SNL), in partnership with Cerebras Systems, announced the launch of the Kingfisher cluster, which will be used as a test platform for the development of AI technologies for national security.

Kingfisher is based on Cerebras CS-3 nodes, which are made on proprietary Wafer Scale Engine accelerators of the third generation (WSE-3). These giant products contain 4 trillion transistors, 900 thousand cores and 44 GB of SRAM memory. The total bandwidth of the built-in memory reaches 21 PB/s, internal interconnect – 214 PB/s.

To date, the Kingfisher platform includes four Cerebras CS-3 nodes, and the final configuration will use eight such units. Cerebras CS-3 nodes with a power of 23 kW each contain life-support systems, power subsystems, Ethernet network interconnect and other components.

Image source: SNL

The deployment of the Cerebras CS-3 cluster is part of the Advanced Simulation and Computing (ASC) program, which is implemented by the US National Nuclear Security Administration (NNSA). We are talking, in particular, about the ASC Artificial Intelligence for Nuclear Deterrence (AI4ND) initiative – artificial intelligence for nuclear deterrence.

Kingfisher is expected to enable the development of large-scale and robust AI models using the secure internal resources of Tri-lab, a group that includes Sandia National Laboratories, Lawrence Livermore National Laboratory (LLNL) and Los Alamos National Laboratory (LANL) within (DOE ).