CORDIC Accelerator 1: Hardware acceleration
CPUs are general purpose computers. A CPU is like a Leatherman: Useful for a wide range of tasks, but a chef wouldn't use a Leatherman to replace her 8" knife.
We can build custom computation architectures that excel at performing specialized computations.
GPUs are specialized for SIMD type computations, which makes them well suited for computer graphics because they can operate simultaneously on huge swaths of pixels. This also makes them appropriate for running neural network matrix operations in parallel.
Any computation can be targeting in order to accelerate it. Perhaps the computation contains parallelizable elements which would otherwise be handled serially by a CPU. Or perhaps sequential computation throughput can be optimized on the dataflow level using techniques like pipelining.
The computation I seek to accelerate for this project is robot coordinate system transforms.
