Cerebras Releases Tungsten for High-Level Wafer-Scale Programming

Unite.AI
Read full story· 5 min
Cerebras Systems launched Tungsten, a high-level programming language and compiler for its Wafer-Scale Engine, enabling scientists to program the 900,000-core chip without hardware expertise. Alongside Tungsten, Cerebras released Aviary, a suite of scientific applications and tutorials demonstrating wafer-scale programming.

What happened

On November 16, 2026, Cerebras Systems introduced Tungsten, a high-level programming language and compiler designed for its Wafer-Scale Engine. The company also launched Aviary, a collection of scientific applications and tutorials that demonstrate how to use the new language on the 900,000-core chip. The release aims to let scientists write programs for the wafer-scale hardware without needing deep hardware knowledge. A plasma simulation written entirely by an AI coding agent has already been run using Tungsten, covering 98% of the wafer.

Key details

  • Tungsten is a high-level language and compiler for the Cerebras Wafer-Scale Engine, which has 900,000 cores.
  • Aviary includes scientific applications such as heat equations, reaction-diffusion stencils, particle transport, and molecular dynamics.
  • A plasma simulation created by a researcher and an AI coding agent advanced 60 million particles at 7,200 time steps per second, achieving 430 billion particle updates per second.
  • The SNAP application reached 1.6 PFLOP/s on one CS-4 system, surpassing performance from 10,800 GPUs on a supercomputer.
  • Tungsten code can run on a single wafer or scale across clusters, with deployments up to 2,048 systems.

Why it matters

Scientists and researchers working with Cerebras hardware can now develop complex scientific applications without needing to manually manage low-level hardware details. This reduces the time and expertise required to use the Wafer-Scale Engine for large-scale simulations and scientific computing. Organizations using Cerebras systems gain the ability to scale their workloads more easily and leverage AI agents to assist in code development, potentially speeding up research and application deployment.


More in Machine Learning