Intel will be relying on Facebook’s expertise for the design of its next-genCopper Lake Xeon processor. Facebook will help the chip giant implement Bfloat 16 for deep learning.
Bfloat 16 is a half-precision FP format that boosts deep learning performance by offering the same level of dynamic range as FP32. Like most FP16 based formats, it improves the efficiency of tasks that don’t require 32-bit precision, such as speech recognition, facial detection, machine translation, etc.
Just like its predecessor, Cascade Lake, Copper Lake will leverage Intel’s 14nm++ node and will be used in the HEDT and server chips. Intel along with Chinese OEM Inspur has developed a new server motherboard design with four Xeon sockets. This can help build stations with 112 cores in a 2U design.
OEMs including Dell, HP, Lenovo, Supermicro, and ZT Systems will be coming up with server motherboards based on this architecture.