NVIDIA Considers Direct-to-Chip Cooling for Next-Gen Rubin Ultra to Address Thermal Issues from Power-Hungry AI Chips

NVIDIA Considers Direct-to-Chip Cooling for Next-Gen Rubin Ultra to Address Thermal Issues from Power-Hungry AI Chips

NVIDIA is reportedly preparing to adopt a groundbreaking cooling solution for its upcoming Rubin Ultra AI series, aiming to address increasing power demands and thermal management challenges.

Transitioning to Microchannel Cover Plates: A New Era for NVIDIA’s Rubin Ultra

As technological advancements continue to escalate power consumption, efficient cooling solutions have become paramount for companies like NVIDIA. As outlined by @QQ_Timmy, NVIDIA is exploring partnerships with cooling solution providers. The goal is to implement ‘direct-to-chip’ cooling using microchannel cold plates in the Rubin Ultra AI graphics processing units (GPUs).This move may mark a significant departure from traditional liquid cooling methods, potentially enabling NVIDIA to maximize performance levels while managing thermals more effectively.

To better grasp the importance of microchannel cover plates (MCCP), it is helpful to draw parallels with direct-die cooling, commonly used by overclocking enthusiasts in contemporary CPUs. This method employs a copper cold plate embedded with microchannels that facilitate coolant circulation, enhancing local convection and significantly reducing thermal resistance between the chip and the fluid. Unlike NVIDIA’s existing cooling techniques, MCCP modifies the liquid-cooler plates directly, allowing for far superior thermal regulation.

Moving beyond the technical elements, it is essential to discuss why NVIDIA is pursuing this innovative technology. The company’s ongoing product schedule demands rapid enhancements, and the evolution from the Blackwell to the Rubin architecture introduces notably increased power consumption, especially in intricate rack-scale deployments. This pressing need compels NVIDIA to explore cutting-edge cooling solutions, as architectural innovations drive the demand for advanced thermal management.

Rubin Ultra NVL576 presentation with specifications including '15 EF FP4 Inference' and '115.2 TB/s CX9 8X.'
Image Credits: NVIDIA

Reports suggest that NVIDIA is in talks with Asia Vital Components, a Taiwanese thermal solutions provider, to engineer the MCCP solution for the Rubin Ultra lineup. However, due to stringent timelines, there may be challenges in pivoting to this advanced solution quickly. Furthermore, Microsoft has recently introduced a microfluidic cooling strategy, akin to MCCP, but with a distinct approach focusing on ‘in-chip cooling.’ This emphasizes the urgent demand across the industry for pioneering cooling technologies.

Source & Images

Leave a Reply

Your email address will not be published. Required fields are marked *