Meta가 오하이오주 Prometheus 데이터센터 캠퍼스의 세부 내용을 공개했으며, 1단계가 이미 가동 중이고 Muse 학습에 활용되고 있습니다.

Meta has detailed progress on its Prometheus data center campus in New Albany, Ohio, which it plans to scale to a gigawatt. The site was originally used as a traditional Facebook data center, but the company said it has now converted the facility into the first phase of its AI-focused supercluster.

In a blog post, Meta said this deployment has already been used to train its Muse foundation models. The campus currently houses hundreds of thousands of GPUs across multiple interconnected buildings, including traditional and AI-optimized data center buildings and rapid deployment structures, all connected. At full scale, Prometheus is planned to comprise roughly one million distributed GPUs.

The data center houses Nvidia GB200 and GB300 chips, with Vera Rubin "arriving soon."

The GB200 is supported by Meta's Catalina rack architecture. The company is deploying "the Ariel variant, a two-rack configuration in which 36 GPUs per rack are interconnected via NVLink to form a unified 72-GPU domain," Meta's engineers said. The Ariel variant of the GB200 NVL72 pairs each Nvidia Blackwell GPU with a Grace CPU and two ConnectX-7 400G NICs at a 1:1 CPU-to-GPU ratio, supporting 800G of I/O per GPU.

For the GB300 deployment, Meta uses its Clemente platform, which can use either Meta's modular Air-Assisted Liquid Cooling (AALC) or Facility Liquid Cooling (FLC).

"GB300 raises HBM capacity per GPU by 1.5x, from 192GB to 288GB, and dense FP4 throughput by a similar factor, while roughly doubling attention (softmax) acceleration," Meta said. "At the platform level, scale-out is no longer the constraint it was on the GB200 NVL72: Nvidia ConnectX-8 SuperNICs deliver 800G per GPU natively, so the capability we deployed early on Ariel becomes the baseline."

At the building level, each Ethernet-based fabric interconnects up to 20,736 GPUs through a high-bandwidth, low-latency multi-layer topology. Building-level fabrics are then connected by the Backend Aggregation (BAG) network, which spans distances from 2 km to over 100 km.

"It is built with Arista 7800-series chassis running deep-buffered Broadcom Jericho3 silicon, delivering multiple Pbps of bandwidth per region pair," Meta said.