The present work is on the implementation and analysis of the PSO (Particle Swarm Optimization)
algorithm, an algorithm that opens possibility for the parallelization in CUDA, using a
technique of the Compute Capability 6.1, shared memory, allowing the decrease of the latency
of communication between the CUDA core and memory. This work aimed to implement the
PSO algorithm so that the result obtained after its execution was preciser in such ways that it
could obtain more precise decimal results, besides improving the execution time of it using the
new architecture of the NVidia GPUs GTX 10 series, Pascal. For the analysis and comparison
of results, it has been used results of the work produced by Daniel Souza in 2014, and in the
obtained results an improvement was reached both in the precision, where it was increased
from 6 decimal places to 16 decimal places, and at run time of the algorithm, which opens the
possibility for applications that require extreme precision, such as medical applications and
biomedicine, can be executed without the execution time being affected.
|
