[Manuel Calavera]

The Hopper GPU, which is primarily aimed at the Datacenter & AI segment, is allegedly taping out soon and will feature an MCM architecture. NVIDIA won't be using an MCM design on its Ada Lovelace GPUs so they will keep the traditional monolithic design.

AMD on the other hand is going on board with offering both MCM and Monolithic chips in its RDNA 3 and CDNA 2 families. The CDNA 2 GPUs will be MCM only while RDNA 3 will feature a mix of MCM and monolithic design as detailed here. It isn't confirmed which process node each company will be relying on but based on speculation, AMD's RDNA 3 & CDNA 2 family is going to be a mix of 6nm & 5nm nodes featuring new and refreshed GPUs while NVIDIA is expected to utilize TSMC's 5nm process node for its Ada Lovelace GPUs though it is not confirmed if those will be fabricated on the N5 or N5P node. Intel, on the other hand, may also rely on TSMC's 6nm process node for its own ARC Alchemist lineup of GPUs which enter mass production later this year and launch in Q1 2022.