[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"aihot-art-99520":3},{"itemId":4,"vertical":5,"category":6,"source":7,"score":8,"title":9,"summary":10,"analysis":11,"url":12,"coverUrl":13,"direction":13,"marketSignal":13,"publishedAt":14},"99520","ai","行业动态","IT之家",59,"DeepSeek R1 推理 AI 测试：英伟达 Vera Rubin NVL72 每兆瓦 Tokens 吞吐量提升 10 倍","英伟达Vera Rubin NVL72在DeepSeek R1推理测试中，每兆瓦Tokens吞吐量提升10倍。","英伟达 Vera Rubin NVL72 在 DeepSeek R1 推理测试中，能效比 Blackwell 提升 10 倍，迈向万亿级部署。\n· 测试对比 Vera Rubin NVL72 与 GB200 NVL72，在相同交互性目标下，每兆瓦 token 吞吐量提升 10 倍。\n· Vera Rubin 已在 CoreWeave、Google Cloud 等平台运行，供应链覆盖 30 国 350+ 工厂。\n· CoreWeave 于 6 月初完成首个 Vera Rubin NVL72 上电验证，系统端到端就绪。\n影响\u002F看点：能效大幅提升将加速推理模型规模化部署，降低 AI 应用成本。","https:\u002F\u002Fwww.ithome.com\u002F0\u002F979\u002F825.htm",null,"2026-07-22 07:34:13"]