Bỏ qua đến nội dung chính
Back to home
AI Tech 2 min read

NVIDIA GB300 NVL72 Achieves 10x Higher Performance Per Watt Than Hopper

NVIDIA GB300 NVL72 systems achieve 10x higher performance per watt than Hopper when running the Kimi K2.6 model, solving a key power constraint for AI factories.

Tier 1 · sources 67% confidence Reviewed
📚 Aggregated from 2 sources X — @nvidia X — @nvidia

NVIDIA has announced that its GB300 NVL72 systems deliver up to 10x higher performance per watt than the previous Hopper architecture when running the Kimi K2.6 model. This breakthrough comes at a critical time when global data centers are facing strict power constraints.

Detailed Developments

NVIDIA made this announcement as technology companies actively seek solutions to optimize energy consumption for AI infrastructure. By running real-world tests of the Kimi K2.6 model on the GB300 NVL72 system, the chip giant demonstrated the significant optimization capabilities of its new architecture. Improving energy efficiency not only cuts operating costs for cloud providers but also addresses power shortages at supercomputing facilities.

Technical & Technology Analysis

Performance per watt is now considered the foundational metric for every AI factory operating in a power-constrained environment. The GB300 NVL72 system leverages high-density interconnect architecture and next-generation Blackwell chips to minimize data transmission loss. The synergy between improved hardware and software optimization for the Kimi K2.6 model enabled the system to achieve this 10x efficiency milestone over its predecessor, Hopper.

Expert Opinions & Insights

According to NVIDIA, performance per watt is the deciding factor for modern data factories. Technical analysts note that sustaining computational growth without causing a surge in power consumption is key to maintaining the momentum of generative AI. The superiority of the GB300 NVL72 over the Hopper line indicates that NVIDIA is actively addressing the semiconductor industry's biggest bottleneck.

Impact & Future

This technological milestone is expected to drive a massive wave of infrastructure upgrades across major data centers. For large model developers like Kimi, energy-efficient operations will lower the cost of AI services, making intelligent applications more accessible to end-users. The trend of chip design focusing heavily on power efficiency will continue to dominate the tech industry in the coming years.