Intel's FalconShores GPU to Be Produced by TSMC with Advanced Processes

TapTechNews July 16th news, Taiwan media Anue钜亨网 reported on the 14th of this month that Intel's next-generation GPU chip for AI and HPC, FalconShores, will be produced by TSMC. It has currently completed Tapeout and will enter mass production by the end of next year.

Specifically, Intel's FalconShores GPU will use TSMC's 3nm and 5nm advanced manufacturing processes, and the HBM integration will also use TSMC's CoWoS-R process.

TapTechNews note:

CoWoS-R is a 2.5D packaging and integration process that completely replaces the silicon interposer layer with the RDL layer and has a lower cost. The structure is shown in the following figure:

The report also pointed out that Intel has planned at least three different grades of chips on the FalconShores product line, aiming at the high, middle and low-end markets respectively to fully compete for AI accelerator orders.

Neither TSMC nor Intel has commented on this market rumor.

According to Intel's previous statements, FalconShores will become the common successor of PonteVecchio data center GPU and Gaudi 3 accelerator, carrying the next-generation Intel Xe architecture and integrating the essence of the Gaudi architecture, and is scheduled to be released in 2025.

Intel will implement a rapid iterative strategy on the FalconShores product line, and the second-generation product is scheduled to be launched in 2026.