每秒110万个token!微软(MSFT.US)和英伟达(NVDA.US)联手刷新AI推理纪录

智通财经
Nov 04, 2025

智通财经APP获悉,微软(MSFT.US)宣布,其Azure ND GB300v6虚拟机在Meta的Llama270B模型上实现了每秒推理速度达110万token的行业新纪录。据悉,Azure ND GB300虚拟机采用英伟达(NVDA.US)的Blackwell Ultra GPU,具体为NVIDIA GB300NVL72系统,配置72个NVIDIA Blackwell Ultra GPU和36个NVIDIA Grace CPU,采用单机架构设计。这款虚拟机专为推理工作负载优化,具有50%的GPU内存提升和16%的热设计功率(TDP)提高。

微软首席执行官萨提亚・纳德拉在社交媒体上表示:“这一成就是我们与英伟达长期合作和在生产规模运行人工智能方面专业知识的结晶。”

资料显示,为了验证性能提升,微软在一个NVIDIA GB300NVL72域下的18台ND GB300v6虚拟机上运行了Llama270B(FP4精度)模型,使用NVIDIA TensorRT-LLM作为推理引擎。微软表示:“一个NVL72机架的Azure ND GB300v6实现了总计110万个token/秒的推理速度。” 这一新记录超越了微软之前在NVIDIA GB200NVL72机架上实现的86.5万个token/秒的成绩。

对此,Signal65的实验室副总裁拉斯・费洛斯指出:“这一里程碑不仅突破了每秒百万token的障碍,还在一个能够满足现代企业动态使用和数据治理需求的平台上实现。” 他补充称,Azure ND GB300相较于上一代NVIDIA GB200在推理性能上提升了27%,而仅增加了17%的功率规格。

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Most Discussed

  1. 1
     
     
     
     
  2. 2
     
     
     
     
  3. 3
     
     
     
     
  4. 4
     
     
     
     
  5. 5
     
     
     
     
  6. 6
     
     
     
     
  7. 7
     
     
     
     
  8. 8
     
     
     
     
  9. 9
     
     
     
     
  10. 10