Thrust CUDA calls are synchronous to each other. You can add an explicit synchronization call cudaDeviceSynchronize() and there won't be a difference in results.