CudaDigitalAnnealing

Code Source

Overall time complexity: O(sweep*dim)

Hardware and docker image information

Used OS: Ubuntu 20.04.3 LTS

kernel: Linux 5.13.0-51-generic

Used image version: nvidia/cuda:10.0-devel-ubuntu18.04

Used image card: NVIDIA GeForce 3090

command used to create container:

docker run -it --gpus all nvidia/cuda:10.0-devel-ubuntu18.04 bash

To use the algorithm:

First of all, generate a cudaDA.so file for the cudaDigitalAnnealing.cu file.

In command line, run the following prompt:

nvcc --compiler-options -fPIC -shared -arch sm_70 --maxrregcount=255 -o ./cudaDA.so cudaDigitalAnnealing.cu

Locate the cudaDA.so file and main.py file, put them in the same folder.

Second, import the algorithm to your code:

from main import DA

Then you can use the algorithm with following code:

algorithm = DA()
algorithm.run()

result:

For a ising problem with 727 binary bits, the algorithm spent 10 seconds to finish 100,000 iterations to find a local minima of the problem.

compare to SA:

the prototype of the algorithm spent 59 seconds to finish 100,000 iterations, compared to python Simulated Annealing algorithm which spent 200 seconds