Panthera is the cycle-accurate software performance model built by Ampere's Architecture & Performance team to study prospective CPU features, evaluate microarchitectural tradeoffs, and guide the development of our high-performance cloud processors.

See the User Guide for information about building and using Panthera and its associated tooling.

Historically, industrial-grade performance models calibrated against production silicon have remained closely guarded trade secrets. We decided to open-source Panthera because we believe the broader computer systems and architecture community stands to benefit immensely from having access to a modern, commercial-caliber modeling infrastructure. Because AmpereOne is actively deployed and accessible across major cloud providers, researchers now have an unprecedented opportunity to study, experiment with, and directly correlate a software model against real, production silicon running in the cloud.

Here is why we are opening Panthera to the world:

For decades, academic computer architecture research and graduate education have relied heavily on academic simulators. While foundational, these tools often diverge from the complex realities and trade-offs of commercial server processors.

Panthera bridges this divide by providing an industrial-strength performance model developed for designing CPU products. Because Ampere processors are widely available in the public cloud, researchers, educators, and practitioners can run workloads on both this model and actual production silicon. This enables the community to validate ideas against realistic microarchitectural baselines, evaluate concepts in an authentic server context, and separate the architectural "wheat from the chaff."

Panthera was built from scratch by CPU architects bringing decades of experience from across the semiconductor industry. The modeling abstractions reflect practical lessons learned while building high-performance CPUs. By sharing this infrastructure alongside our companion tools like catscan, we aim to foster open collaboration across industry and academia, planting the seeds for better modeling practices and novel microarchitectural ideas.

Low-background steel refers to steel forged before atmospheric nuclear testing, free of background radiation and prized as a baseline for sensitive scientific instruments.

In that same spirit, Panthera represents low-background steel for computer systems. It is an end-to-end, industrial-grade performance model conceived, designed, and hand-crafted entirely by human architects before the era of generative AI. This repository represents over 8 years of work and 6000 individual git commits by 45 people. By making this human-authored codebase public, we hope it serves as a trusted, high-watermark reference for human engineers and future automated hardware-modeling tools alike.

Panthera is released as a public reference implementation under a permissive open-source license. While Ampere does not provide commercial support or active feature roadmaps for this public repository, we actively welcome the community to explore the codebase, adapt it for research, and fork it for novel architectural exploration.

Configuration files in the repository use code names for the SoC product and the custom cores within them. Here is a quick decoder ring between the external marketing names and the engineering names.

Panthera models several generations of Ampere processors. The AmpereOne (Siryn) model configuration contains our original custom core designed for high performance and low-power cloud computing: Furiosa. The Furiosa microarchitecture notes describe it in more detail.

The product floorplan gallery maps the floorplans for AmpereOne, AmpereOne X, AmpereOne MX, and AmpereOne Aurora to their SoC and core code names. Each of these cores builds upon the previous micro-architecture. Likewise, our model uses 'delta' configurations which build upon the previous generation's.