RoCE network scaling capacity
Company: CoreWeave
The claim, verbatim
CoreWeave's modular RoCE network design can extend the two-tier architecture to approximately 128,000 GPUs
Source (primary)
What It Takes to Bring Up a Multi-Rack NVIDIA Vera Rubin NVL72 Cluster - CoreWeave (-, news_article)
View cached copy (2026-09-19)Live source ↗
Quote: “Our current modular design increases spine capacity within each plane and can extend the two-tier architecture to approximately 128,000 GPUs”
How we checked this
Checked on September 24, 2026. The cited source supports every part of this claim.
The post says CoreWeave's two-tier RoCE fabric for Vera Rubin can be extended, under its current modular design, to approximately 128,000 GPUs.
Confirmed in the source:
- CoreWeave's Vera Rubin RoCE network uses a two-tier architecture
- CoreWeave's current modular design can extend that two-tier architecture to approximately 128,000 GPUs
What we did: Read our cached copy of the publisher (https://www.coreweave.com/blog/what-it-takes-to-bring-up-a-multi-rack-nvidia-vera-rubin-nvl72-cluster) in full (17,292 characters, retrieved September 19, 2026) and checked each assertion in the claim against it.
Additional evidence
confirms What It Takes to Bring Up a Multi-Rack NVIDIA Vera Rubin NVL72 Cluster - CoreWeave
Quote: “Our current modular design increases spine capacity within each plane and can extend the two-tier architecture to approximately 128,000 GPUs”