Heterogeneous Computing in the Edge
|
To Internet Backbone |
|
|
|
CISC / RISC CPU Servers CISC / RISC CPU Servers
CISC / RISC CPU Servers
GPU Accelerators GPU Accelerators GPU Accelerators
|
Cloud Interconnec t |
FPGA Farms
FPGA Farms
FPGA Farms
Tensor Proc . Units
Tensor Proc . Units Tensor Proc . Units
|
Cloud Data Center |
CISC / RISC Server Modules
GPU Accelerator Modules
Storage Modules
Backplane / Interconnec t
Interface Modules
FPGA Farm Modules
Te ns or Processing Modules
Edge Computer Node
Fig . 1 : Heterogeneous Computing in the Cloud - Edge Hierarchy .
Edge computing nodes in common deployment today typically use a homogenous computing base , that is all of the processing done on an edge node runs on the same type of CPU . CISC / RISC processors are by far the most popular solutions . Smaller edge nodes may have a single core CPU chip providing their processing power , often using X86 or ARM architectures . Larger edge nodes include multicore processors , with between two and about 32 X86 , ARM , or RISC-V cores , or include multiple CPU chips of the same type .
Processor vendors such as Intel 3 , AMD 4 , and the ARM ecosystem 5 have defined conventional CISC / RISC processor chips that are optimized for these edge computing applications , by virtue of their specialized data path architectures , buses , power dissipation profiles , I / O structures , memory and storage subsystems , physical size , and environmental specifications . Those processor vendors provide hardware reference designs to aid implementers in the application of these processor chips to edge node applications .
3
4
5
IIC Journal of Innovation - 45 -