Intel is working with digital infrastructure provider Vertiv on a new liquid cooling solution for its Gaudi3 AI accelerator chipset.
Set to launch in 2024, the Gaudi3 processor is designed for the most demanding data centre workloads, AI training, and inferencing, positioning it as Intel's contender against the industry-leading Nvidia AI technology.
The Intel Gaudi3 AI accelerator will enable both liquid-cooled and air-cooled servers, supported by Vertiv pumped two-phase (P2P) cooling infrastructure.
The cooling solution is crucial for AI applications and high-performance computing. Performing intense computations on massive amounts of data leads to increased power consumption and heat generation. Cooling solutions help dissipate this heat, ensuring that the chip operates at its best performance without the risk of overheating.
John Nieman, SVP global thermal line of businesses at Vertiv, says the company is expanding its liquid cooling portfolio to support big tech companies such as Intel.
Liquid cooling also provides a more eco-friendly and cost-efficient solution compared to traditional cooling methods, although it is more complex to install.
“To support increasing thermal design power and heat flux for next-generation accelerators, Intel has worked with Vertiv and other ecosystem partners to enable a cooling solution that will help customers meet critical sustainability goals,” says Dr Devdatta Kulkarni, principal engineer of this project at Intel.
The microprocessor giant introduced Gaudi2 in 2022, aiming to rival Nvidia's A100 GPUs as the preferred choice for enterprises training large language models. The acquisition of Israel-based Habana for $2 billion in late 2019 paved the way for this development.