python 详解如何使用GPU大幅提高效率_Python

python 详解如何使用GPU大幅提高效率

2022-02-26 00:22微小冷 Python

CuPy是一个开源矩阵库，使用NVIDIA CUDA加速。CuPy使用Python提供GPU加速计算。CUPY使用CUDA相关库，包括 CuBLAS、CUDNN、Curand、CuoSver、CuPaSeSE、Cufft和NCCL，以充分利用GPU架构

cupy我觉得可以理解为cuda for numpy，安装方式pip install cupy，假设

				?

									import numpy as np

									import cupy as cp

那么对于np.XXX一般可以直接替代为cp.XXX。

其实numpy已经够快了，毕竟是C写的，每次运行的时候都会尽其所能地调用系统资源。为了验证这一点，我们可以用矩阵乘法来测试一下：在形式上通过多线程并发、多进程并行以及单线程的方式，来比较一下numpy的速度和对资源的调度情况，代码为

				?

									# th_pr_array.py

									from threading import Thread

									from multiprocessing import Process

									from time import time as Now

									import numpy as np

									import sys

									N = 3000

									def MatrixTest(n,name,t):

									    x = np.random.rand(n,n)

									    x = x@x

									    print(f"{name} @ {t} : {Now()-t}")

									def thTest():

									    t = Now()

									    for i in range(5):

									        Thread(target=MatrixTest,args=[N,f'th{i}',t]).start()

									def prTest():

									    t = Now()

									    for i in range(5):

									        Process(target=MatrixTest,args=[N,f'pr{i}',t]).start()

									if __name__=="__main__":

									    if sys.argv[1]=="th":

									        thTest()

									    elif sys.argv[1]=="pr":

									        prTest()

									    else:

									        t = Now()

									        for i in range(5):

									            MatrixTest(N,"single",t)

运行结果为

(base) E:\Documents\00\1108>python th_pr_numpy.py th
th0 @ 1636357422.3703225 : 15.23965334892273
th1 @ 1636357422.3703225 : 17.726242780685425
th2 @ 1636357422.3703225 : 19.001763582229614
th3 @ 1636357422.3703225 : 19.06676197052002
th4 @ 1636357422.3703225 : 19.086761951446533

(base) E:\Documents\00\1108>python th_pr_numpy.py pr
pr3 @ 1636357462.4170427 : 4.031360864639282
pr0 @ 1636357462.4170427 : 4.55387806892395
pr1 @ 1636357462.4170427 : 4.590881824493408
pr4 @ 1636357462.4170427 : 4.674877643585205
pr2 @ 1636357462.4170427 : 4.702877759933472

(base) E:\Documents\00\1108>python th_pr_numpy.py single
single @ 1636357567.8899782 : 0.36359524726867676
single @ 1636357567.8899782 : 0.8137514591217041
single @ 1636357567.8899782 : 1.237830400466919
single @ 1636357567.8899782 : 1.683635950088501
single @ 1636357567.8899782 : 2.098794937133789

所以说在numpy中就别用python内置的并行和并发了，反而会称为累赘。而且这么一比更会印证numpy的强大性能。

但在cupy面前，这个速度会显得十分苍白，下面连续5次创建5000x5000的随机矩阵并进行矩阵乘法，

				?

									#np_cp.py

									import numpy as np

									import cupy as cp

									import sys

									from time import time as Now

									N = 5000

									def testNp(t):

									    for i in range(5):

									        x = np.random.rand(N,N)

									        x = x@x

									    print(f"np:{Now()-t}")

									def testCp(t):

									    for i in range(5):

									        x = cp.random.rand(N,N)

									        x = x@x

									    print(f"cp:{Now()-t}")

									if __name__ == "__main__":

									    t = Now()

									    if sys.argv[1] == 'np':

									        testNp(t)

									    elif sys.argv[1]=='cp':

									        testCp(t)