TensorFlow 使用 GPUs

支持的设备

在一套标准的计算机系统上通常有多个计算设备. TensorFlow 支持 CPU 和 GPU 这两种设备. 我们用指定字符串来标识这些计算设备. 比如:

• "/cpu:0": 机器中的 CPU
• "/gpu:0": 机器中的 GPU, 如果你有一个的话.
• "/gpu:1": 机器中的第二个 GPU, 以此类推...

如果一个 TensorFlow 的 operation 中兼有 CPU 和 GPU 的实现, 当这个算子被指派设备时, GPU 有优先权. 比如matmul中 CPU 和 GPU kernel 函数都存在. 那么在 cpu:0 和 gpu:0 中, matmul operation 会被指派给 gpu:0 .

记录设备指派情况

为了获取 operations 和 Tensor 被指派到哪个设备上运行, 用 log_device_placement 新建一个 session, 并设置为 True.

# 新建一个 graph.
a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')
b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')
c = tf.matmul(a, b)

# 新建session with log_device_placement并设置为True.
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))

# 运行这个 op.
print(sess.run(c))

输出:

Device mapping:
/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: GeForce GTX 750 Ti, pci bus id: 0000:04:00.0, compute capability: 5.0
MatMul: (MatMul): /job:localhost/replica:0/task:0/device:GPU:0
b: (Const): /job:localhos2019-03-24 17:27:19.202619: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\direct_session.cc:297] Device mapping:
/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: GeForce GTX 750 Ti, pci bus id: 0000:04:00.0, compute capability: 5.0

2019-03-24 17:27:19.205594: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\placer.cc:874] MatMul: (MatMul)/job:localhost/replica:0/task:0/device:GPU:0
t/replica:0/task:0/device:GPU:0
2019-03-24 17:27:19.207181: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\placer.cc:874] b: (Const)/job:localhost/replica:0/task:0/device:GPU:0
2019-03-24 17:27:19.208401: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\placer.cc:874] a: (Const)/job:localhost/replica:0/task:0/device:GPU:0
a: (Const): /job:localhost/replica:0/task:0/device:GPU:0
[[22. 28.]
 [49. 64.]]

手工指派设备

如果你想手工指派设备, 用 with tf.device 创建一个设备环境, 这个环境下的 operation 都统一运行在环境指定的设备上.

# 新建一个graph.
with tf.device('/cpu:0'):
  a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')
  b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')
c = tf.matmul(a, b)

# 新建session with log_device_placement并设置为True.
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))

# 运行这个op.
print(sess.run(c))

现在 a 和 b 操作都被指派给了 cpu:0.

Device mapping:
2019-03-24 17:30:26.190956: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\direct_session.cc:297] Device mapping:
/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: GeForce GTX 750 Ti, pci bus id: 0000:04:00.0, compute capability: 5.0
/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: GeForce GTX 750 Ti, pci bus id: 0000:04:00.0, compute capability: 5.0

MatMul: (MatMul): /job:localhost/replica:0/task:0/device:GPU:0
b: (Const): /job:localhost/replica:0/task:0/device:CPU:0
a: (Const): /job:localhost/replica:0/task:0/device:CPU:0
2019-03-24 17:30:26.193904: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\placer.cc:874] MatMul: (MatMul)/job:localhost/replica:0/task:0/device:GPU:0
2019-03-24 17:30:26.195145: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\placer.cc:874] b: (Const)/job:localhost/replica:0/task:0/device:CPU:0
2019-03-24 17:30:26.196196: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\placer.cc:874] a: (Const)/job:localhost/replica:0/task:0/device:CPU:0
[[22. 28.]
 [49. 64.]]

在多GPU系统里使用单一GPU

如果你的系统里有多个 GPU, 那么 ID 最小的 GPU 会默认使用. 如果你想用别的 GPU, 用下面的方法显式的声明你的偏好:

# 新建一个graph.
with tf.device('/gpu:2'):
  a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')
  b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')
c = tf.matmul(a, b)

# 新建session with log_device_placement并设置为True.
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))

# 运行这个op.
print(sess.run(c))

如果指定的设备不存在, 你会收到 InvalidArgumentError 错误提示:

InvalidArgumentError (see above for traceback): Cannot assign a device for operation 'b': Operation was explicitly assigned to /device:GPU:2 but available devices are [ /job:localhost/replica:0/task:0/device:CPU:0, /job:localhost/replica:0/task:0/device:GPU:0 ]. Make sure the device specification refers to a valid device.
    [[Node: b = Const[dtype=DT_FLOAT, value=Tensor<type: float shape: [3,2] values: [1 2][3]...>, _device="/device:GPU:2"]()]]

为了避免出现你指定的设备不存在这种情况, 在创建的 session 里把参数 allow_soft_placement 设置为 True, 这样 tensorFlow 会自动选择一个存在并且支持的设备来运行 operation.

# 新建一个 graph.
with tf.device('/gpu:2'):
  a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')
  b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')
  c = tf.matmul(a, b)

# 新建 session with log_device_placement 并设置为 True.
sess = tf.Session(config=tf.ConfigProto(
      allow_soft_placement=True, log_device_placement=True))

# 运行这个 op.
print(sess.run(c))

使用多个 GPU

如果想让 TensorFlow 在多个 GPU 上运行, 你可以建立 multi-tower 结构, 在这个结构 里每个 tower 分别被指配给不同的 GPU 运行. 比如:

# 新建一个 graph.
c = []
for d in ['/gpu:0', '/gpu:1']:
  with tf.device(d):
    a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3])
    b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2])
    c.append(tf.matmul(a, b))
with tf.device('/cpu:0'):
  sum = tf.add_n(c)

# 新建session with log_device_placement并设置为True.
sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True, log_device_placement=True))

# 运行这个op.
print(sess.run(sum))

输出:

Device mapping:
2019-03-24 17:37:25.516312: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\direct_session.cc:297] Device mapping:
/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: GeForce GTX 750 Ti, pci bus id: 0000:04:00.0, compute capability: 5.0
/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: GeForce GTX 750 Ti, pci bus id: 0000:04:00.0, compute capability: 5.0

MatMul_1: (MatMul): /job:localhost/replica:0/task:0/device:GPU:0
MatMul: (MatMul): /job:localhost/replica:0/task:0/device:GPU:0
2019-03-24 17:37:25.519521: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\placer.cc:874] MatMul_1: (MatMul)/job:localhost/replica:0/task:0/device:GPU:0
2019-03-24 17:37:25.520587: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\placer.cc:874] MatMul: (MatMul)/job:localhost/replica:0/task:0/device:GPU:0
2019-03-24 17:37:25.522257: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\placer.cc:874] AddN: (AddN)/job:localhost/replica:0/task:0/device:CPU:0
2019-03-24 17:37:25.523393: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\placer.cc:874] Const_3: (Const)/job:localhost/replica:0/task:0/device:GPU:0
2019-03-24 17:37:25.524425: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\placer.cc:874] Const_2: (Const)/job:localhost/replica:0/task:0/device:GPU:0
2019-03-24 17:37:25.526241: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\placer.cc:874] Const_1: (Const)/job:localhost/replica:0/task:0/device:GPU:0
2019-03-24 17:37:25.527728: IAddN: (AddN): /job:localhost/replica:0/task:0/device:CPU:0
Const_3: (Const): /job:localhost/replica:0/task:0/device:GPU:0
Const_2: (Const): /job:localhost/replica:0/task:0/device:GPU:0
Const_1: (Const): /job:localhost/replica:0/task:0/device:GPU:0
Const: (Const): /job:localhost/replica:0/task:0/device:GPU:0
 C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\placer.cc:874] Const: (Const)/job:localhost/replica:0/task:0/device:GPU:0
[[ 44.  56.]
 [ 98. 128.]]

cifar10 tutorial 这个例子很好的演示了怎样用GPU集群训练.