(written for windows) compile with nvcc -O3 -arch=sm_86 --use_fast_math cuda_vector_add.cu run with ./a.exe 1000000000 the argument is the length of arrays can't run with arrays much larger than that