CoreWeave brings NVIDIA Vera Rubin systems to production

CoreWeave announced availability of NVIDIA Vera Rubin NVL72 systems with Spectrum-X 102.4T Ethernet networking at CoreWeave Fully Connected in San Francisco, extending its co-engineered NVIDIA cloud platform into production for agentic workloads. Cognition, the applied AI lab behind Devin, is the first customer running production workloads on Vera Rubin.

Cognition benchmarked Vera Rubin against a GB200 NVL72 baseline using a real-world software engineering workload built from FrontierCode tasks. Early tests showed up to a 4.8x increase in total token throughput for SWE-2 inference workloads over GB200 NVL72, gains aimed at faster real-time code generation and more responsive multistep reasoning for Devin.

CoreWeave will also add NVIDIA Vera CPU and said its deployment packs 128 CPUs and 11,264 cores in a single rack, enough for more than 11,000 concurrent environments at one core each. Testing showed more than 3x faster agent sandbox startup times on NVIDIA Vera CPUs and a 1.7x performance gain on Terminal-Bench across all passing tasks.

CoreWeave Forge combines Weights & Biases, OpenPipe post-training expertise and the open source marimo notebook project for continuous model and agent improvement across models, frameworks and clouds. Canva, Capital One and MasterClass are among first builders on Forge, while Ennoble Care selected CoreWeave for clinical AI inference serving about 50,000 high-need Medicare patients across 15 states.