Launched last March, the Apple M1 Ultra SoC uses a die-to-die connection interface that allows you to connect two M1 Max chips, achieving practically double the performance, without the compromises that usually occur when connecting a chip. to another through the lines on the motherboard. The custom packaging architecture is called UltraFusion, which DigiTimes colleagues have offered more accurate information about in the past, which you can find in our previous article.
Photo Credit: TSMC / Tom Wassick / Twitter According to DigiTimes, the Cupertino company used TSMC's 2.5D CoWoS-S (chip-on-wafer-on-substrate with silicon interposer) packaging process, but apparently this is not correct. In fact, according to a presentation given by TSMC itself at the International Symposium on 3D IC and Heterogeneous Integration, Apple uses Integrated Fan-Out (InFO) technology with Local Silicon Interconnect (LSI) and a Redistribution Layer (RDL).
Photo Credit: TSMC / Tom Wassick / Twitter We remind you that M1 Ultra is almost 8 times faster than M1 and offers a maximum of 20 CPU cores, divided into 14 high-performance cores and 6 high-performance cores efficiency, a maximum of 128GB of RAM (64GB per chip), double the channels and a bandwidth of up to 800GB / s. The graphics sector can count on a GPU with 64 cores, while the Neural Engine is equipped with 32 cores for managing AI loads.
Photo Credit: TSMC / Tom Wassick / Twitter According to DigiTimes, the Cupertino company used TSMC's 2.5D CoWoS-S (chip-on-wafer-on-substrate with silicon interposer) packaging process, but apparently this is not correct. In fact, according to a presentation given by TSMC itself at the International Symposium on 3D IC and Heterogeneous Integration, Apple uses Integrated Fan-Out (InFO) technology with Local Silicon Interconnect (LSI) and a Redistribution Layer (RDL).
Photo Credit: TSMC / Tom Wassick / Twitter We remind you that M1 Ultra is almost 8 times faster than M1 and offers a maximum of 20 CPU cores, divided into 14 high-performance cores and 6 high-performance cores efficiency, a maximum of 128GB of RAM (64GB per chip), double the channels and a bandwidth of up to 800GB / s. The graphics sector can count on a GPU with 64 cores, while the Neural Engine is equipped with 32 cores for managing AI loads.