$ cat thrust_async.cu

#include <iostream>
#include <chrono>
#include <thrust/random.h>
#include <thrust/device_vector.h>
#include <thrust/transform.h>
#include <thrust/iterator/counting_iterator.h>
#include <thrust/iterator/constant_iterator.h>

class Timer
{
public:
    Timer() : beg_(clock_::now()) {}
    void reset() { beg_ = clock_::now(); }
    double elapsed() const {
        return std::chrono::duration_cast<second_>
            (clock_::now() - beg_).count(); }

private:
    typedef std::chrono::high_resolution_clock clock_;
    typedef std::chrono::duration<double, std::ratio<1> > second_;
    std::chrono::time_point<clock_> beg_;
};

struct prg
{
    float a, b;

    __host__ __device__
    prg(float _a=0.f, float _b=1.f) : a(_a), b(_b) {};

    __host__ __device__
        float operator()(const unsigned int n) const
        {
            thrust::default_random_engine rng;
            thrust::uniform_real_distribution<float> dist(a, b);
            rng.discard(n);

            return dist(rng);
        }
};


int main(void)
{
    const int N = 1 << 20;

    {
    thrust::device_vector<float> numbers(N);
    thrust::device_vector<float> results(N);

    thrust::counting_iterator<unsigned int> index_sequence_begin(0);
    thrust::constant_iterator<unsigned int> keys_in(0);
    thrust::device_vector<unsigned int> keys_out(N);

    Timer timer;
    timer.reset();

    thrust::transform(index_sequence_begin,
            index_sequence_begin + N,
            numbers.begin(),
            prg(1.f,2.f));

    double t1 = timer.elapsed();

    auto output = thrust::reduce_by_key(keys_in, keys_in+N, numbers.begin(), keys_out.begin(), results.begin());

    double t2 = timer.elapsed();
    cudaDeviceSynchronize();
    double t3 = timer.elapsed();

    std::cout << "thrust::reduce_by_key took " << 1e3*(t2-t1) << "ms" << std::endl;
    std::cout << "cudaDeviceSynchronize took " << 1e3*(t3-t2) << "ms" << std::endl;
    }

    cudaDeviceReset();

    return 0;
}

$ nvcc -std=c++11 -arch=sm_52 -o thrust_async thrust_async.cu

$ ./thrust_async
thrust::reduce_by_key took 1.64952ms
cudaDeviceSynchronize took 0.004666ms