In the AI era, power is the ultimate constraint, and every AI factory operates within a hard limit. This makes performance per watt—the rate at which power is…

NVIDIA DevTalk serves as a vibrant community hub where developers can engage in discussions, seek assistance, and collaborate on projects involving NVIDIA hardware and software. Developers can tap into the collective expertise of the NVIDIA developer community, sharing insights, troubleshooting issues, and exploring best practices for GPU programming and AI development. Additionally, DevTalk provides a platform for developers to showcase their projects, receive feedback, and network with peers, fostering collaboration and knowledge exchange within the NVIDIA ecosystem.

NVIDIA Developer

NVIDIA outlines how its GPU architectures and AI factory software maximize performance per watt to increase token throughput and revenue within fixed power envelopes. Across six architecture generations, NVIDIA claims a 1,000,000x improvement in inference throughput per megawatt. Key highlights include: Blackwell Ultra GB300 NVL72 delivering up to 50x higher throughput per megawatt and 35x lower token cost versus Hopper for DeepSeek-R1; the upcoming Vera Rubin platform achieving up to 10x higher inference throughput per megawatt versus Blackwell; 100% liquid cooling enabling 1.1 PUE; and the NVIDIA DSX system enabling AI factories to operate up to 30% more GPUs within the same power envelope. The post also covers efficiency gains in chip manufacturing via cuLitho and cuEST, GPU-accelerated EDA, and how operators should track revenue per megawatt as the core business metric.

Scaling Token Factory Revenue and AI Efficiency by Maximizing Performance per Watt

Compounding performance per watt across NVIDIA GPU architectures

Building for efficiency with extreme co-design