Intel's Ponte Vecchio data center GPU uses Foveras chiplet technology and CXL interconnects for advanced 7nm packaging.

Intel is developing a powerful 7nm Xe-based GPU codenamed "Ponte Vecchio," named after an old bridge in Florence, Italy. LinkedIn recruitment advertisements referencing Arctic Sound and Ponte Vecchio-based solutions suggest ongoing hiring for the project. These GPUs are slated for Project Aurora, an exascale supercomputer from Intel expected to go online in 2021.

According to Videocardz, up to six Ponte Vecchio GPUs will power the Aurora supercomputer alongside two Sapphire Rapids Xeon processors. The GPUs will be connected using CXL (Compute Express Link) technology with a OneAPI software stack and will utilize Foveros packaging technology. Despite not being targeted at gamers, the Ponte Vecchio GPUs are designed to deliver ultra-high cache, high memory bandwidth, and high double-precision FP throughput.

Intel's broader Xe GPU family is planned to operate across multiple segments including HPC/Exascale, DL/Training, Cloud GFX, Media Transcore Analytics, Workstation, Gaming, PC Mobile and Ultra Mobile. Intel planned to reveal further details about the 7nm Xe GPU "Ponte Vecchio" on November 17, 2019.

The appearance of "Ponte Vecchio" in multiple LinkedIn job listings alongside the previously known codename "Arctic Sound" indicates that Ponte Vecchio is a derivative of the Arctic Sound GPU family, which was targeted directly at the Data Center. This connection suggests Intel has been working on this project for an extended period.

Intel's plans for consumer gaming GPUs remain unclear, though the company is expected to launch discrete GPUs in late-2020. Whether this initial launch will be a gaming-focused GPU or data-centric product is uncertain, though Intel may debut with a lower-powered training vehicle such as the DG1 before ramping up to more powerful offerings that could compete with NVIDIA's TITANs and challenge AMD in the entry-level segment.