Intel’s Next-Gen Falcon Shores GPU Will Not Be Air-Cooled and Will Use 1500 W

Though it will consume a lot of power, Intel’s next Falcon Shores GPU is expected to be a powerhouse for AI and high-performance computing (HPC) tasks. It is anticipated that the processor, which combines the Gaudi and Ponte Vecchio successors into a single GPU, will draw an incredible 1500 W of power, which is more than even Nvidia’s powerful B200 accelerator, which draws 1000 W. To guarantee the Falcon Shores GPU runs effectively and safely, cutting-edge cooling technologies will be needed due to its enormous power consumption. The technology that Intel has been pushing for power-hungry data center hardware, liquid cooling, or even full immersion liquid cooling, may be used by Intel’s partners. The strong performance promises of the Falcon Shores GPU come at a cost: a considerable power draw.In comparison to its Ponte Vecchio devices, Intel promises to deliver five times greater performance per watt as well as five times more memory and bandwidth.

Given that the existing Open Accelerator Module (OAM) 2.0 limits out at roughly 1000 W, Intel could need to create proprietary hardware modules or a new OAM specification in order to enable such high power levels. The Falcon Shores GPU, which will be based on Intel’s next-generation Xe graphics architecture, is scheduled for availability in 2025. With the support of Intel’s extensive oneAPI software development environment, it hopes to become a significant participant in the AI accelerator industry. Although the 1500 W power consumption will undoubtedly draw criticism, Intel is hoping that the Falcon Shores GPU’s purportedly excellent performance will make it a compelling choice for AI and HPC users prepared to make a significant investment in reliable cooling infrastructure. The market for ultra-high-end accelerators is rapidly expanding, and the HPC accelerator segment requires a Ponte Vecchio replacement.