CUDA_Nearest_Neighbors

The aim of this project is to measure the speedup obtained for the algorithm K Nearest Neighbors when implemented in OpenMP/CUDA with respect to the C++ sequential version.

System configuration

• Intel Core i7-8750H @ 2.20Ghz (up to 4.10Ghz with Turbo Boost)
• 16 GiB RAM
• NVIDIA GeForce GTX 1050 Ti (Notebook)
• CUDA 10.0

Datasets

Sift and siftsmall datasets can be found in http://corpus-texmex.irisa.fr, where they are called ANN_SIFT1M and ANN_SIFT10K respectively.

Experiments

Download SIFT dataset

• wget ftp://ftp.irisa.fr/local/texmex/corpus/sift.tar.gz
• mkdir data
• tar -xf sift.tar.gz -C data

Download this repo

• git clone https://github.com/fedem96/Nearest_Neighbors.git

Run the experiments

• mkdir experiments
• cd Nearest_Neighbors
• nvcc -O3 src/main.cu --gpu-architecture=compute_61 -Xcompiler -fopenmp -o run
• ./run -d ../data -e ../experiments

Results

OpenMP

CUDA

Comparison