NVIDIA CORP · News & developments
Vera Rubin moves from NVIDIA’s roadmap to a customer’s production workload
CoreWeave announced availability of NVIDIA Vera Rubin NVL72 on its cloud, with Cognition as the first customer running production workloads on the system. CoreWeave said Cognition’s engineers measured up to 4.8 times the total token throughput on SWE-2 inference workloads versus a GB200 NVL72 baseline. That is a customer-run, workload-specific benchmark; the announcement did not disclose deployment volume or NVIDIA revenue from the rollout.
Why this matters
A production workload is a more useful proof point than a launch announcement: it shows a cloud partner has moved the system into customer use, and gives prospective buyers a real workload comparison. But the reported throughput gain belongs to one benchmark, not a general performance guarantee. For NVIDIA, the open question is whether production use spreads enough to drive meaningful shipments; CoreWeave’s announcement gives no scale or revenue figures to answer it.
Written with AI from the linked sources and reviewed by a SageNoodle editor. How we work.
Explore NVIDIA CORP research →