Unexpected behavior with more than 160 cuda blocks

Basically when I run over 160 blocks at once the function seems to have sometimes executed twice on one kernel, when it should only be executing once. I'm using a gtx 1050 I thought I had like 900 cores or something?

The function should return all 118's in the array but it returns one 1063 and then the rest unchanged 13's.

I am new to cuda (and C++ in general) and I am fuzzy on pointers.

#include "cuda_runtime.h"  #include "device_launch_parameters.h"  #include <iostream>  #include <stdio.h>    __host__ __device__ unsigned int random(unsigned int &seed) {      seed = seed * 9 + 1;      return seed;  }    __global__ void rndKernel(unsigned int *seeds)  {      random(seeds[blockIdx.x]);  }    int main()  {      const unsigned int size = 161;      unsigned int* dev_seeds;      unsigned int* seeds;        cudaMalloc((void**)&dev_seeds, sizeof(unsigned int) * size);        seeds = new unsigned int[size];      for (int i = 0; i < size; i++) {          seeds[i] = 13;      }        cudaMemcpy(dev_seeds, seeds, sizeof(unsigned int) * size, cudaMemcpyHostToDevice);        rndKernel <<< size, 1 >>> (dev_seeds);        cudaMemcpy(seeds, dev_seeds, sizeof(unsigned int) * size, cudaMemcpyDeviceToHost);        cudaFree(dev_seeds);        for (int i = 0; i < size; i++) {          std::cout << seeds[i] << '\n';      }        free(seeds);      return 0;  }    

https://stackoverflow.com/questions/65588992/unexpected-behavior-with-more-than-160-cuda-blocks January 06, 2021 at 09:34AM