Vera Rubin NVL72 inference tests show up to 7x better token throughput per MW vs. Blackwell on a 1.6T DeepSeek model, above Huang's 3x claim for 1T-3T LLMs (Bryan Shan/SemiAnalysis)
SemiAnalysis says Nvidia’s Vera Rubin NVL72 beat Blackwell by a wider efficiency margin than Jensen Huang had claimed.
The note points to inference tests on a 1.6T-parameter DeepSeek model, where Vera Rubin NVL72 delivered up to 7x better token throughput per megawatt versus Blackwell. Huang had claimed a 3x gain for 1T-3T LLMs, so the reported test result lands above that benchmark. The item is framed around power-normalized throughput, not raw chip speed. Source: Techmeme's note.
The note points to inference tests on a 1.6T-parameter DeepSeek model, where Vera Rubin NVL72 delivered up to 7x better token throughput per megawatt versus Blackwell. Huang had claimed a 3x gain for 1T-3T LLMs, so the reported test result lands above that benchmark. The item is framed around power-normalized throughput, not raw chip speed. Source: Techmeme's note.
score 6