www.magazine-industry-usa.com
05
'24
Written on Modified on
Leonardo News
LEONARDO AND GEMATEG COLLABORATE FOR THE THERMAL MANAGEMENT OF NEXT-GENERATION CHIPS
DaTEG, the solution developed by GemaTEG, will increase the efficiency of next-generation chips in Leonardo's Artificial Intelligence (AI) infrastructure by 30%.
www.leonardo.com
Leonardo and GemaTEG have initiated a partnership to reduce the energy costs of next-generation chips used in Artificial Intelligence (AI) infrastructures. GemaTEG has developed DaTEG, the first integrated and modular active thermal management system. Leonardo will deploy this innovative solution in its High-Performance Computing (HPC) systems to reduce the energy footprint of datacenters and increase their efficiency.
The partnership involves GemaTEG's laboratories and Leonardo's Chieti datacenter. This collaboration will explore a wide range of applications in the most relevant and impactful technological areas, with plans to expand to the HPC davinci-1, Leonardo's supercomputer. The goal is to achieve an average efficiency increase of 30% by reducing energy consumption and increasing chip performance in some of the configurations within Leonardo's AI infrastructures. This initiative aims to combine innovation and sustainability in the AI sector.
DaTEG allows precise control of operating conditions, by adjusting the cooling power to the actual chip loads, with a "localized cooling" approach. This provides a dynamic regulation of the temperature of each individual chip. Next-generation chips, such as Central Processing Units (CPUs), Graphics Processing Units (GPUs), and accelerators, driven by performance demands, generate unprecedented amounts of heat, increasing from 300-400 watts to over 1000 watts. Conventional cooling technologies struggle to dissipate this heat level. Overheated chips respond by slowing down computational cycles and reducing performance up to 50% less.
The benefits of adopting the DaTEG solution are numerous. The computing power of each chip is increased because it runs consistently in optimal conditions. Each GPU in the cluster is independently controlled through with localized cooling, allowing the system to address and eliminate bottlenecks from chip overheating. Total energy consumption is minimized with customized thermal management strategies that automatically decide where and how much to cool.
DaTEG”s adaptive thermal management system increases computing power within the same infrastructure while reducing energy consumption associated with cooling. This innovation transcends AI industry limitations by enhancing performance while meeting sustainability objectives.
www.leonardo.com