One thread per byte across a 5-billion-byte buffer

An 80 GB GPU holds a 5,000,000,000-byte buffer, zeroed beforehand. The kernel is launched with one thread per byte, 256 threads per block, and 19,531,250 blocks. After it completes, the buffer is copied back and inspected. What does it contain?

bump.cu
__global__ void bump(unsigned char* data, size_t n) {
    size_t i = blockIdx.x * blockDim.x + threadIdx.x;
    if (i < n) data[i] += 1;
}

size_t n = 5000000000ULL;
unsigned blocks = (unsigned)((n + 255) / 256);   // 19,531,250
bump<<<blocks, 256>>>(d_data, n);

46 of 46 in GPU / CUDA