---
title: "Inside Prometheus: Meta’s Journey to a Gigawatt-Scale Supercluster"
description: "Meta’s Prometheus supercluster in New Albany, Ohio, is a gigawatt-scale AI infrastructure project designed to train frontier models like Muse Wonder. The first phase is complete, comprising hundreds of thousands of GPUs across multiple interconnected buildings, with a target of roughly 1 million GPUs at full scale. Prometheus deploys multiple NVIDIA GPU platforms, including Catalina (GB200 NVL72) with 13.4 TB of HBM3e and Clemente (GB300 NVL72) with 20.7 TB of HBM3e, with Vera Rubin NVL72 arriving soon. The cluster uses a dedicated Back-End Ethernet network for GPU-to-GPU communication, and employs closed-loop liquid cooling and power management via Dynamo Dimmer to maintain high goodput. By focusing on stability, Meta achieved over 85% goodput for pre-training and over 90% for reinforcement learning runs, with a mean time between interruptions of ~47k GPU hours, later improved to over 54k GPU hours."
---

# Inside Prometheus: Meta’s Journey to a Gigawatt-Scale Supercluster

[Meta](https://yomu.fyi/company/meta) · By Kevin Lee, Yashar Bayani, Ahmed Sharif, Yee Jiun Song · Oct 19, 2026

**Type:** Explainer

## Summary

Meta’s Prometheus supercluster in New Albany, Ohio, is a gigawatt-scale AI infrastructure project designed to train frontier models like Muse Wonder. The first phase is complete, comprising hundreds of thousands of GPUs across multiple interconnected buildings, with a target of roughly 1 million GPUs at full scale. Prometheus deploys multiple NVIDIA GPU platforms, including Catalina (GB200 NVL72) with 13.4 TB of HBM3e and Clemente (GB300 NVL72) with 20.7 TB of HBM3e, with Vera Rubin NVL72 arriving soon. The cluster uses a dedicated Back-End Ethernet network for GPU-to-GPU communication, and employs closed-loop liquid cooling and power management via Dynamo Dimmer to maintain high goodput. By focusing on stability, Meta achieved over 85% goodput for pre-training and over 90% for reinforcement learning runs, with a mean time between interruptions of ~47k GPU hours, later improved to over 54k GPU hours.

## Context

Meta is undertaking its most ambitious AI infrastructure project to date, building a gigawatt-scale supercluster called Prometheus in New Albany, Ohio, to train frontier models like Muse Wonder.

## Approach / What changed

Building Prometheus in phases using NVIDIA GPU platforms (GB200 NVL72, GB300 NVL72, and upcoming Vera Rubin NVL72), a dedicated Back-End Ethernet network for GPU-to-GPU communication, closed-loop liquid cooling, and Dynamo Dimmer power management to achieve high goodput.

## Takeaways

- Prometheus achieved over 85% goodput for pre-training and over 90% for reinforcement learning runs by focusing on system stability and reliability.
- The supercluster uses a dedicated Back-End Ethernet network for GPU-to-GPU communication, designed to match the hierarchy of training parallelism.
- Power management via Dynamo Dimmer gradually adjusts GPU power limits to handle large power transients without interrupting training.

## FAQ

### What goodput did Prometheus achieve for training runs?

Prometheus achieved over 85% goodput for pre-training runs and over 90% for reinforcement learning runs of Muse Wonder, with a mean time between interruptions of ~47k GPU hours, later improved to over 54k GPU hours.

### How does Prometheus handle power fluctuations?

Prometheus uses Dynamo Dimmer, a cluster-wide power management framework that gradually adjusts GPU power limits as the cluster nears its electrical ceiling, supported by rack-level battery backup and facility-level battery energy storage systems.

### What GPU platforms does Prometheus use?

Prometheus currently deploys NVIDIA GB200 NVL72 (Catalina) and GB300 NVL72 (Clemente) platforms, with Vera Rubin NVL72 arriving soon.

**Tags:** [AI](https://yomu.fyi/topic/ai), [Data Centers](https://yomu.fyi/topic/data-centers), [GPU](https://yomu.fyi/topic/gpu), [Infrastructure](https://yomu.fyi/topic/infrastructure), [NVIDIA](https://yomu.fyi/topic/nvidia)

- Source: [Meta](https://engineering.fb.com/2026/10/19/data-center-engineering/inside-prometheus-meta-gigawatt-scale-supercluster)
- Source URL: https://engineering.fb.com/2026/10/19/data-center-engineering/inside-prometheus-meta-gigawatt-scale-supercluster
- Ingested by Yomu: 2026-10-19T15:00:11.134Z

[Read original post](https://engineering.fb.com/2026/10/19/data-center-engineering/inside-prometheus-meta-gigawatt-scale-supercluster)