FLOP Benchmark with numpy

flop_benchmark.py

#!/usr/bin/env python3

import time
import numpy as np

N = 4096
A = np.random.randn(N, N).astype(np.float32)
B = np.random.randn(N, N).astype(np.float32)

# number of multiplications
flop = N*N*2*N

for i in range(10):
  start_time = time.monotonic()
  C = A @ B
  end_time = time.monotonic()
  time_diff = end_time - start_time

  print(f"{flop / time_diff * 1e-12:.2f} TFLOP/S")

On a Thinkpad T440p with an i7-4800MQ I get:

0.25 TFLOP/S
0.25 TFLOP/S
0.26 TFLOP/S
0.25 TFLOP/S
0.19 TFLOP/S
0.21 TFLOP/S
0.26 TFLOP/S
0.26 TFLOP/S
0.20 TFLOP/S
0.21 TFLOP/S