Artificial Intelligence1 min read

Frostbite: A new CPU architecture discovered, designed and verified by AI

By · Published by Everything Blog

In short

The architecture relies on an interaction net execution model where computational steps remain strictly local and values are consumed only once. In rigorous testing across 25 workloads, the design delivered a 4.59x performance-per-watt advantage over a modern silicon-proven dual-issue superscalar core. An end-to-end automated workflow co-designed the entire hardware and software stack, discovering novel primitives from the instruction set down to the physical layout. This expansion of earlier automated accelerator research demonstrates the viability of machine-driven exploration for general-purpose processors.

Key points

  • Replacing shared mutable states with interaction nets removes the heavy area and power ov…: Replacing shared mutable states with interaction nets removes the heavy area and power overhead typically caused by traditional cache coherence protocols.
  • Automated co-design of both hardware and software stacks enables the discovery of highly…: Automated co-design of both hardware and software stacks enables the discovery of highly efficient architectural primitives beyond conventional human engineering methods.
  • The developers expect this end-to-end machine-driven methodology to expand into datacente…: The developers expect this end-to-end machine-driven methodology to expand into datacenter processors, AI accelerators, and networking silicon.

Sources checked by our analysis pipeline: 1

Abstract

In our previous post, Redwood, we showed that given an architecture, our AI systems could implement, verify, and deploy an inference accelerator in record time. However, similar to other scientific domains, we believe that AI systems, given the correct freedom and guardrails, can co-design software and hardware together, making optimizations across the stack and opening a design space so large it was previously inaccessible. We set out to answer the question: “Could our AI systems & tools lead to a novel hardware discovery that radically outperforms human-made designs?”

Today, we're introducing Frostbite, a general-purpose CPU with a new instruction set built from the ground up. Frostbite exploits the properties of interaction nets, namely that every step is local, every dependency is explicit, and every value has exactly one owner, to reach a step-function gain in CPU performance that has long stagnated. On a set of 25 initially chosen workloads, including 15 held out until the final evaluation, we achieved a 4.59x performance-per-watt multiple when compared to a modern, silicon-proven superscalar (dual-issue, in-order) CPU core.

Our AI systems autonomously iterated…

Artificial Intelligence