GPU使用

GPU

In [1]:

import mxnet as mx
from mxnet import nd
from mxnet.gluon import nn

In [2]:

mx.cpu(),mx.gpu(),mx.gpu(1)

Out[2]:

(cpu(0), gpu(0), gpu(1))

In [3]:

x = nd.array([1,2,3])
x

Out[3]:

[1. 2. 3.]
<NDArray 3 @cpu(0)>

In [4]:

x.context

Out[4]:

cpu(0)

In [5]:

!nvidia-smi

 

Wed Nov 28 15:42:11 2018       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 384.90                 Driver Version: 384.90                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  TITAN X (Pascal)    Off  | 00000000:02:00.0 Off |                  N/A |
| 33%   57C    P2    58W / 250W |   3467MiB / 12189MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   1  TITAN X (Pascal)    Off  | 00000000:03:00.0 Off |                  N/A |
| 39%   63C    P2    63W / 250W |   1280MiB / 12189MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   2  TITAN X (Pascal)    Off  | 00000000:83:00.0 Off |                  N/A |
| 44%   75C    P2    86W / 250W |   1320MiB / 12189MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   3  TITAN X (Pascal)    Off  | 00000000:84:00.0 Off |                  N/A |
| 43%   74C    P2    84W / 250W |   1719MiB / 12189MiB |      8%      Default |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0      6577      C   python                                       209MiB |
|    0      8489      C   python                                      2329MiB |
|    0      9813      C   python                                       211MiB |
|    0     29538      C   python                                       173MiB |
|    1     17615      C   python                                       591MiB |
|    1     31208      C   python                                       679MiB |
|    2      6577      C   python                                       547MiB |
|    2      9813      C   python                                       763MiB |
|    3      6577      C   python                                       547MiB |
|    3      9813      C   python                                       763MiB |
|    3     15351      C   python                                       399MiB |
+-----------------------------------------------------------------------------+

In [6]:

a = nd.array([1,2,3],ctx=mx.gpu())

In [7]:

a

Out[7]:

[1. 2. 3.]
<NDArray 3 @gpu(0)>

In [8]:

b = nd.random.uniform(shape=(2,3),ctx=mx.gpu(1))
b

Out[8]:

[[0.59119    0.313164   0.76352036]
 [0.9731786  0.35454726 0.11677533]]
<NDArray 2x3 @gpu(1)>

In [9]:

a.context

Out[9]:

gpu(0)

In [10]:

b.context

Out[10]:

gpu(1)

In [11]:

y = x.copyto(mx.gpu())
y

Out[11]:

[1. 2. 3.]
<NDArray 3 @gpu(0)>

In [12]:

z = x.as_in_context(mx.gpu())
z

Out[12]:

[1. 2. 3.]
<NDArray 3 @gpu(0)>

In [13]:

y.as_in_context(mx.gpu()) is y

Out[13]:

True

In [14]:

y.copyto(mx.gpu()) is y

Out[14]:

False

In [15]:

z

Out[15]:

[1. 2. 3.]
<NDArray 3 @gpu(0)>

In [16]:

y

Out[16]:

[1. 2. 3.]
<NDArray 3 @gpu(0)>

In [17]:

(z + 2).exp()

Out[17]:

[ 20.085537  54.59815  148.41316 ]
<NDArray 3 @gpu(0)>

In [18]:

(z+2).exp()*y

Out[18]:

[ 20.085537 109.1963   445.2395  ]
<NDArray 3 @gpu(0)>

 

gluon的GPU计算

In [19]:

net = nn.Sequential()

In [20]:

net.add(nn.Dense(1))

In [21]:

net.initialize(ctx=mx.gpu())

In [22]:

net(y)

Out[22]:

[[0.0068339 ]
 [0.01366779]
 [0.02050169]]
<NDArray 3x1 @gpu(0)>

In [23]:

net[0].weight.data()

Out[23]:

[[0.0068339]]
<NDArray 1x1 @gpu(0)>

In [ ]: