Why my Xeon E5-2640 v3 2.60GHz runs twice faster than my Xeon E5-4627 v3 2.6GHz

Question

I am struggling with the performance of the newly bought machine with the Xeon E5-4627 CPU. My experiment codes ran almost twice as slow on Xeon E5-4627 v3 CPU than the same code ran on Xeon E5-2640 v3 processor. I did some simple benchmark on these two machines. Here is the code

long N = 1 << 24;
double s = 0;
for (long i = 0; i < N; i++)
   s += std::cos(i % 360);

I compiled the code on the machine with E5-4627 as g++ -std=c++11 -O3 -o test test.cpp and ran the same binary execution file in these two machines. The E5-4627 cost about 0.95s while the E5-2640 cost about 0.36s.

Both machines are with Ubuntu 4.8.5 and the g++-4.8 was used.