Next-Gen AI Processors Expected to Consume Up to 15,360 Watts: A Shift Toward Advanced Immersion and Embedded Cooling Technologies

The power consumption of AI GPUs is projected to surge, with Nvidia’s next-gen GPUs potentially reaching thermal design power (TDP) between 6,000W and 9,000W. KRIST researchers anticipate TDPs could hit as high as 15,360W by 2035. Cooling demands will escalate from traditional high-performance air cooling to advanced methods like immersion cooling and embedded cooling solutions. Upcoming GPUs, such as Nvidia’s Blackwell Ultra and Rubin Ultra, indicate a trend, with TDPs of 1,400W and 3,600W, respectively. Innovations like thermal transmission lines (TTLs) and fluidic thermal vias (F-TSVs) will be essential for effective heat management. Researchers suggest that while immersion cooling might suffice until 2032, future architectures will necessitate integrated cooling systems, employing techniques that enhance heat dissipation and thermal stability. By 2038, advanced cooling solutions will include double-sided interposers and improved GPU designs to manage thermal challenges in AI computing.

Source link