50 Years of Computer Architecture: From Mainframe CPUs to Neural-Network TPUs - Crossminds
CrossMind.ai logo
50 Years of Computer Architecture: From Mainframe CPUs to Neural-Network TPUs
Sep 20, 2020
|
1195 views
Details
This talk was presented by David Patterson, Turing Award Laureate and professor at the University of California, Berkeley. It reviews a half-century of computer architecture: We start with the IBM System 360, which in 1964 introduced the concept of “binary compatibility”. Next, came the idea of the “dominant microprocessor architecture”, for which the early candidate was the Intel 432 which was shortly replaced by the emergency introduction of the Intel 80x86 in 1978. However, for the next 20 years, the Reduced Instruction Set Computers (RISC) became dominant. Then, the Very Long Instruction Word (VLIW) HP/Intel Itanium architecture was heralded as their replacement in 2001, but instead the role was usurped by AMD’s introduction of the 64 bit 80x86. Thus, while the 80x86 dominated the PC-Era, RISCs have led thereafter, currently with 20B shipped annually (versus 0.4B 80x86s). Since the ending of Moore’s Law and Dennard scaling has stalled performance of general-purpose microprocessors, domain- specific computer architectures are the only option left. An early example of this trend introduced by Google in 2015 is the Tensor Processing Unit (TPU) for cloud-based deep neural networking .
Comments
loading...
Reactions (0) | Note
    📝 No reactions yet
    Be the first one to share your thoughts!
loading...
Recommended