Accelerated Computing

GPU
Custom Chips
FPGA

GPU

Distributed load across thousands of cores.
Refer Working of GPU and gpu programming paradigms

Custom Silicon

Fastest, but Extremely difficult and expensive to implement

Steps involved

1. develop design
2. simulate
3. prototype
4. silicon conversion

FPGA: programmable hardware

Custom hardware

More on this in FPGA

Ways to program FPGA

1. Hardware design - HDL - core electronics and VLSI
2. High level synthesis (HLS) - C to RTL (Resistor Transistor Logic)
3. OpenCL based abstraction