Float16 is much slower than Float32 and Float64 in numpy

user_9814710 0
Viewed 4247

I was trying to run a code snippet which looks like,

import numpy as np
import time

def estimate_mutual_info(X, neurons, bins = 5):
    xy = np.histogram2d(X, neurons, bins)[0]
    x = np.histogram(X, bins)[0]
    y = np.histogram(neurons, bins)[0]
    ent_x = -1 * np.sum( x / np.sum(x) * np.log( x / np.sum(x)))
    ent_y = -1 * np.sum( y / np.sum(y) * np.log( y / np.sum(y)))
    ent_xy = -1 * np.sum( xy / np.sum(xy) * np.log( xy / np.sum(xy)))
    return (ent_x + ent_y - ent_xy)

tic = time.time()
X = np.random.rand(12000, 1200)
Y = np.random.rand(12000, 10)
for j in Y.T:
    mi = 0
    for i in range(X.shape[1]):
        mi += estimate_mutual_info(X.T[i], j, bins = 2)
    print(mi)
toc = time.time()
print(str(toc - tic)+" seconds")

To increase the speed, I used float16, hoping to see some improvement, but float16 was much slower than float32 and float64. 

X = np.random.rand(12000, 1200).astype('float16')
Y = np.random.rand(12000, 10).astype('float16')

changing them to float16 results in execution time of 84.57 seconds, whereas float64 and float32 executed for 36.27 seconds and 33.25 seconds respectively. I am not sure, what causes this poor performance for flaot16. My processor is 64 bit, using python3.7 and numpy-1.16.2. I don't think 64 bit processor treats all 16 bit, 32 bit and 64 bit indifferent. Any correction and insight is much appreciated.

numpy python performance
2 Answers

The most likely explanation is that your processor does not natively support FP16 arithmetic, so it's all being done in software, which is, of course, much slower.

In general, consumer Intel processors don't support FP16 operations.

user_110154270

it is happening because there is no equivalent of float16 in c.

since python is based on c, as there is no equivalent of that in c, numpy created a method to perform for float16.

( float is a 32 bit IEEE 754 single precision Floating Point Number1 bit for the sign, (8 bits for the exponent, and 23* for the value), i.e. float has 7 decimal digits of precision)

Becacause of this( process making equivalent to work on float16) float16 is slower than float32 or float64

user_50862550
Related