Cognition puts Nvidia's Vera Rubin to work, reports up to 4.8x token throughput

Cognition, Scott Wu's coding-agent company, ran its SWE-2 workload on CoreWeave's first production Vera Rubin cluster, comparing it with GB200 in a vendor-led test.

By · Published · Updated

Cognition puts Nvidia's Vera Rubin to work, reports up to 4.8x token throughput — Scott Wu's coding-agent company ran its SWE-2 workload on CoreWeave's first production Vera Rubin cluster, comparing it with GB200 in a vendor-led test.

Cognition, Scott Wu's coding-agent company, ran its SWE-2 workload on CoreWeave's first production Vera Rubin cluster, comparing it with GB200 in a vendor-led test.