• Wion
  • /Technology
  • /Nvidia says its next AI superchip is 10 times more efficient than Blackwell

Nvidia says its next AI superchip is 10 times more efficient than Blackwell

Nvidia says its next AI superchip is 10 times more efficient than Blackwell

Nvidia says its next AI superchip is 10 times more efficient than Blackwell

Story highlights

Nvidia has unveiled new performance details for its Vera Rubin AI platform, claiming up to 10 times better token-per-watt efficiency than Grace Blackwell. The system combines CPUs and GPUs to power advanced AI workloads as competition with AMD intensifies.

Nvidia has offered its clearest look yet at Vera Rubin, the AI platform it hopes will power the next generation of artificial intelligence. During a technical briefing at its headquarters in Santa Clara, California, the company revealed fresh performance figures for the new chip system ahead of AMD's annual AI event. Nvidia says the Vera Rubin NVL72 platform can process 10 times more AI tokens per watt than its current Grace Blackwell system, while delivering significantly higher memory bandwidth and easier deployment.

The announcement highlights a major shift in Nvidia's strategy. The company no longer wants to be seen only as the world's leading GPU maker it also wants to become a supplier of complete AI computing systems.

Nvidia is betting on CPUs too

Add WION as a Preferred Source

Graphics Processing Units (GPUs) remain the backbone of AI training, but modern AI agents also rely heavily on Central Processing Units (CPUs) to manage data, networking and software operations.

Vera Rubin reflects that change.

Each NVL72 rack combines 72 Rubin GPUs with 36 Vera CPUs, creating an integrated platform designed for large AI workloads. Nvidia is also planning to sell the Vera CPU separately, expanding its push into a market traditionally dominated by AMD and Intel.

Trending Stories

"We're on a roadmap to crank out new architectures not just GPUs but CPUs," Nvidia Vice President Ian Buck said during the briefing.

Faster, cooler and easier to install

Nvidia says Vera Rubin introduces several practical improvements for data centres.

The company has redesigned the platform to use 100 per cent liquid cooling, reducing energy needed for cooling. It has also dramatically reduced cabling, allowing server racks to be installed or replaced in minutes instead of hours.

Another key improvement is memory performance. Nvidia claims Vera Rubin offers nearly three times the memory bandwidth of Grace Blackwell, an important advantage as AI models continue to grow in size. The company also revealed that OpenAI is already testing a Vera Rubin rack as an early customer.

A crucial launch in the AI chip race

Vera Rubin arrives at a critical moment for Nvidia. The company is competing aggressively with AMD, which is expected to showcase its Helios AI platform, while cloud providers including Microsoft, Oracle, Meta and Amazon continue investing billions of dollars in AI infrastructure.

Nvidia also faces pressure to execute flawlessly after earlier reports of overheating issues affecting some Blackwell server configurations. According to Nvidia, Vera Rubin remains on schedule for full production in the second half of this year. As AI models become larger and more capable, the competition is no longer just about building the fastest GPU. Companies now need complete AI systems that combine processing power, networking, memory efficiency and energy savings and Nvidia is betting Vera Rubin will keep it ahead in that race.

About the Author

Abhinav Yadav

Abhinav is a versatile and adaptive journalist who covers defence, space, and technology for WION. He specialises in breaking down complex subjects into clear, engaging stories tha...Read More

Trending Topics