Original research brief
NVIDIA and CoreWeave Move Vera Rubin NVL72 from Training to Production
CoreWeave announces NVIDIA Vera Rubin NVL72 availability with Spectrum-X 102.4T Ethernet. Cognition reports up to 4.8x token throughput for SWE-2…
What this piece is grounded in
01
CoreWeave announced availability of NVIDIA Vera Rubin NVL72 systems with Spectrum-X 102.4T Ethernet networking.
02
Cognition saw Vera Rubin NVL72 deliver up to a 4.8x increase in total token throughput for SWE-2 inference workloads over GB200 NVL72.
03
CoreWeave achieved more than 3x faster agent sandbox startup times on NVIDIA Vera CPUs.
04
CoreWeave's deployment of Vera puts 128 CPUs and 11,264 cores in a single rack.
01 / FIELD NOTE
What Changed
At CoreWeave Fully Connected in San Francisco, CoreWeave announced availability of NVIDIA Vera Rubin NVL72 systems with Spectrum-X 102.4T Ethernet networking, according to NVIDIA. Cognition, the applied AI lab behind Devin, is the first customer running production workloads on Vera Rubin. CoreWeave also announced NVIDIA Vera CPU, described as the first CPU built for AI agents, and launched CoreWeave Forge, a connected environment for training, evaluating and improving models and agents. CoreWeave received its first Vera Rubin NVL72 production racks earlier this month and stood up a production cluster for Cognition in days.
02 / FIELD NOTE
Why It Matters for AI Work
Cognition benchmarked Vera Rubin's inference performance against a GB200 NVL72 baseline using a real-world software engineering workload sampled from FrontierCode. In early tests, Cognition saw Vera Rubin NVL72 deliver up to a 4.8x increase in total token throughput for SWE-2 inference workloads over GB200 NVL72. For Devin, those gains mean faster real-time code generation and more responsive multistep reasoning. CoreWeave's deployment of Vera CPU puts 128 CPUs and 11,264 cores in a single rack, enough for more than 11,000 concurrent environments at one core each. In testing, CoreWeave achieved more than 3x faster agent sandbox startup times on NVIDIA Vera CPUs.
03 / FIELD NOTE
What You Can Verify
CoreWeave announced availability of NVIDIA Vera Rubin NVL72 on CoreWeave Cloud, making it one of the first cloud providers to deliver the platform in customers' hands. Capacity can be operated through CoreWeave Kubernetes Service, SUNK, CoreWeave Mission Control, CoreWeave Sandboxes and CoreWeave Inference. On Terminal-Bench, CoreWeave saw a 1.7x performance gain on Vera CPU across all passing tasks. CoreWeave Forge unifies Weights & Biases, post-training expertise from OpenPipe and the open source marimo notebook project. Canva, Capital One and MasterClass are among the first companies building on Forge.
04 / FIELD NOTE
Editorial Interpretation: Practical Implications
The reported 4.8x token throughput gain for SWE-2 inference suggests that teams running agentic coding workloads could see meaningful speedups in real-time code generation, based on Cognition's early tests. The 3x faster sandbox startup times on Vera CPU point to potential efficiency gains for reinforcement learning and model evaluation, which require thousands of isolated environments. The launch of CoreWeave Forge, unifying tools like Weights & Biases and marimo, addresses the historically fragmented loop between production behavior and training runs. These are early results from one customer and vendor testing; broader production experience will determine general applicability.
Questions readers ask
What is the NVIDIA Vera Rubin NVL72 system?
NVIDIA Vera Rubin NVL72 is a system announced by CoreWeave as available on CoreWeave Cloud with Spectrum-X 102.4T Ethernet networking. It is designed for AI workloads, and Cognition is the first customer running production workloads on it. Early tests showed up to 4.8x higher token throughput for SWE-2 inference over GB200 NVL72.
What is CoreWeave Forge?
CoreWeave Forge is a connected environment for training, evaluating and improving models and agents on NVIDIA accelerated computing. It unifies Weights & Biases, post-training expertise from OpenPipe and the open source marimo notebook project. It stays open across models, frameworks and clouds, with Canva, Capital One and MasterClass among first companies building on it.