tensorflow:Multiple GPUs

深度学习theano/tensorflow多显卡多人使用问题集

tensorflow中使用指定的GPU及GPU显存

Using GPUs

petewarden/tensorflow_makefile

tf_gpu_manager/manager.py

多GPU运行Deep Learning 和 并行Deep Learning（待续）

---

Multiple GPUs

---

---

1. 终端执行程序时设置使用的GPU

---

如果电脑有多个GPU，tensorflow默认全部使用。如果想只使用部分GPU，可以设置CUDA_VISIBLE_DEVICES。在调用python程序时，可以使用

CUDA_VISIBLE_DEVICES=1 python my_script.py #只使用GPU1CUDA_VISIBLE_DEVICES=0,1 python my_script.py #使用GPU0,GPU1

Environment Variable Syntax      ResultsCUDA_VISIBLE_DEVICES=1           Only device 1 will be seenCUDA_VISIBLE_DEVICES=0,1         Devices 0 and 1 will be visibleCUDA_VISIBLE_DEVICES="0,1"       Same as above, quotation marks are optionalCUDA_VISIBLE_DEVICES=0,2,3       Devices 0, 2, 3 will be visible; device 1 is maskedCUDA_VISIBLE_DEVICES=""          No GPU will be visible

---

2. python代码中设置使用的GPU

---

如果要在python代码中设置使用的GPU，可以使用下面的代码

import osos.environ["CUDA_VISIBLE_DEVICES"] = "2"

---

3. 设置tensorflow使用的显存大小

---

定量设置显存

默认tensorflow是使用GPU尽可能多的显存。可以通过下面的方式，来设置使用的GPU显存：

gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.7)sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))    

上面分配给tensorflow的GPU显存大小为：GPU实际显存*0.7。
可以按照需要，设置不同的值，来分配显存。

按需设置显存

上面的只能设置固定的大小。如果想按需分配，可以使用allow_growth参数（参考网址：http://blog.csdn.net/cq361106306/article/details/52950081）：

gpu_options = tf.GPUOptions(allow_growth=True)sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))   

---

4. 使用多个 GPU

---

如果你想让 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:2', '/gpu:3']:  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(log_device_placement=True))# 运行这个op.print sess.run(sum)

你会看到如下输出:

Device mapping:/job:localhost/replica:0/task:0/gpu:0 -> device: 0, name: Tesla K20m, pci busid: 0000:02:00.0/job:localhost/replica:0/task:0/gpu:1 -> device: 1, name: Tesla K20m, pci busid: 0000:03:00.0/job:localhost/replica:0/task:0/gpu:2 -> device: 2, name: Tesla K20m, pci busid: 0000:83:00.0/job:localhost/replica:0/task:0/gpu:3 -> device: 3, name: Tesla K20m, pci busid: 0000:84:00.0Const_3: /job:localhost/replica:0/task:0/gpu:3Const_2: /job:localhost/replica:0/task:0/gpu:3MatMul_1: /job:localhost/replica:0/task:0/gpu:3Const_1: /job:localhost/replica:0/task:0/gpu:2Const: /job:localhost/replica:0/task:0/gpu:2MatMul: /job:localhost/replica:0/task:0/gpu:2AddN: /job:localhost/replica:0/task:0/cpu:0[[  44.   56.] [  98.  128.]]

---

5. 如何实现multi_gpu_model函数

---

def multi_gpu_model(num_gpus=1):  grads = []  for i in range(num_gpus):    with tf.device("/gpu:%d"%i):      with tf.name_scope("tower_%d"%i):        model = Model(is_training, config, scope)        # 放到collection中，方便feed的时候取        tf.add_to_collection("train_model", model)        grads.append(model.grad) #grad 是通过tf.gradients(loss, vars)求得        #以下这些add_to_collection可以直接在模型内部完成。        # 将loss放到 collection中， 方便以后操作        tf.add_to_collection("loss",model.loss)        #将predict放到collection中，方便操作        tf.add_to_collection("predict", model.predict)        #将 summary.merge op放到collection中，方便操作        tf.add_to_collection("merge_summary", model.merge_summary)        # ...  with tf.device("cpu:0"):    averaged_gradients = average_gradients(grads)# average_gradients后面说明    opt = tf.train.GradientDescentOptimizer(learning_rate)    train_op=opt.apply_gradients(zip(average_gradients,tf.trainable_variables()))  return train_op

---

6. cifar10 tutorial-cifar10_multi_gpu_train.py

---

code 见 models/tutorials/image/cifar10/

# Copyright 2015 The TensorFlow Authors. All Rights Reserved.## Licensed under the Apache License, Version 2.0 (the "License");# you may not use this file except in compliance with the License.# You may obtain a copy of the License at##     http://www.apache.org/licenses/LICENSE-2.0## Unless required by applicable law or agreed to in writing, software# distributed under the License is distributed on an "AS IS" BASIS,# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.# See the License for the specific language governing permissions and# limitations under the License.# =============================================================================="""A binary to train CIFAR-10 using multiple GPUs with synchronous updates.Accuracy:cifar10_multi_gpu_train.py achieves ~86% accuracy after 100K steps (256epochs of data) as judged by cifar10_eval.py.Speed: With batch_size 128.System        | Step Time (sec/batch)  |     Accuracy--------------------------------------------------------------------1 Tesla K20m  | 0.35-0.60              | ~86% at 60K steps  (5 hours)1 Tesla K40m  | 0.25-0.35              | ~86% at 100K steps (4 hours)2 Tesla K20m  | 0.13-0.20              | ~84% at 30K steps  (2.5 hours)3 Tesla K20m  | 0.13-0.18              | ~84% at 30K steps4 Tesla K20m  | ~0.10                  | ~84% at 30K stepsUsage:Please see the tutorial and website for how to download the CIFAR-10data set, compile the program and train the model.http://tensorflow.org/tutorials/deep_cnn/"""from __future__ import absolute_importfrom __future__ import divisionfrom __future__ import print_functionfrom datetime import datetimeimport os.pathimport reimport timeimport numpy as npfrom six.moves import xrange  # pylint: disable=redefined-builtinimport tensorflow as tfimport cifar10FLAGS = tf.app.flags.FLAGStf.app.flags.DEFINE_string('train_dir', '/tmp/cifar10_train',                           """Directory where to write event logs """                           """and checkpoint.""")tf.app.flags.DEFINE_integer('max_steps', 1000000,                            """Number of batches to run.""")tf.app.flags.DEFINE_integer('num_gpus', 1,                            """How many GPUs to use.""")tf.app.flags.DEFINE_boolean('log_device_placement', False,                            """Whether to log device placement.""")def tower_loss(scope, images, labels):  """Calculate the total loss on a single tower running the CIFAR model.  Args:    scope: unique prefix string identifying the CIFAR tower, e.g. 'tower_0'    images: Images. 4D tensor of shape [batch_size, height, width, 3].    labels: Labels. 1D tensor of shape [batch_size].  Returns:     Tensor of shape [] containing the total loss for a batch of data  """  # Build inference Graph.  logits = cifar10.inference(images)  # Build the portion of the Graph calculating the losses. Note that we will  # assemble the total_loss using a custom function below.  _ = cifar10.loss(logits, labels)  # Assemble all of the losses for the current tower only.  losses = tf.get_collection('losses', scope)  # Calculate the total loss for the current tower.  total_loss = tf.add_n(losses, name='total_loss')  # Attach a scalar summary to all individual losses and the total loss; do the  # same for the averaged version of the losses.  for l in losses + [total_loss]:    # Remove 'tower_[0-9]/' from the name in case this is a multi-GPU training    # session. This helps the clarity of presentation on tensorboard.    loss_name = re.sub('%s_[0-9]*/' % cifar10.TOWER_NAME, '', l.op.name)    tf.summary.scalar(loss_name, l)  return total_lossdef average_gradients(tower_grads):  """Calculate the average gradient for each shared variable across all towers.  Note that this function provides a synchronization point across all towers.  Args:    tower_grads: List of lists of (gradient, variable) tuples. The outer list      is over individual gradients. The inner list is over the gradient      calculation for each tower.  Returns:     List of pairs of (gradient, variable) where the gradient has been averaged     across all towers.  """  average_grads = []  for grad_and_vars in zip(*tower_grads):    # Note that each grad_and_vars looks like the following:    #   ((grad0_gpu0, var0_gpu0), ... , (grad0_gpuN, var0_gpuN))    grads = []    for g, _ in grad_and_vars:      # Add 0 dimension to the gradients to represent the tower.      expanded_g = tf.expand_dims(g, 0)      # Append on a 'tower' dimension which we will average over below.      grads.append(expanded_g)    # Average over the 'tower' dimension.    grad = tf.concat(axis=0, values=grads)    grad = tf.reduce_mean(grad, 0)    # Keep in mind that the Variables are redundant because they are shared    # across towers. So .. we will just return the first tower's pointer to    # the Variable.    v = grad_and_vars[0][1]    grad_and_var = (grad, v)    average_grads.append(grad_and_var)  return average_gradsdef train():  """Train CIFAR-10 for a number of steps."""  with tf.Graph().as_default(), tf.device('/cpu:0'):    # Create a variable to count the number of train() calls. This equals the    # number of batches processed * FLAGS.num_gpus.    global_step = tf.get_variable(        'global_step', [],        initializer=tf.constant_initializer(0), trainable=False)    # Calculate the learning rate schedule.    num_batches_per_epoch = (cifar10.NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN /                             FLAGS.batch_size)    decay_steps = int(num_batches_per_epoch * cifar10.NUM_EPOCHS_PER_DECAY)    # Decay the learning rate exponentially based on the number of steps.    lr = tf.train.exponential_decay(cifar10.INITIAL_LEARNING_RATE,                                    global_step,                                    decay_steps,                                    cifar10.LEARNING_RATE_DECAY_FACTOR,                                    staircase=True)    # Create an optimizer that performs gradient descent.    opt = tf.train.GradientDescentOptimizer(lr)    # Get images and labels for CIFAR-10.    images, labels = cifar10.distorted_inputs()    batch_queue = tf.contrib.slim.prefetch_queue.prefetch_queue(          [images, labels], capacity=2 * FLAGS.num_gpus)    # Calculate the gradients for each model tower.    tower_grads = []    with tf.variable_scope(tf.get_variable_scope()):      for i in xrange(FLAGS.num_gpus):        with tf.device('/gpu:%d' % i):          with tf.name_scope('%s_%d' % (cifar10.TOWER_NAME, i)) as scope:            # Dequeues one batch for the GPU            image_batch, label_batch = batch_queue.dequeue()            # Calculate the loss for one tower of the CIFAR model. This function            # constructs the entire CIFAR model but shares the variables across            # all towers.            loss = tower_loss(scope, image_batch, label_batch)            # Reuse variables for the next tower.            tf.get_variable_scope().reuse_variables()            # Retain the summaries from the final tower.            summaries = tf.get_collection(tf.GraphKeys.SUMMARIES, scope)            # Calculate the gradients for the batch of data on this CIFAR tower.            grads = opt.compute_gradients(loss)            # Keep track of the gradients across all towers.            tower_grads.append(grads)    # We must calculate the mean of each gradient. Note that this is the    # synchronization point across all towers.    grads = average_gradients(tower_grads)    # Add a summary to track the learning rate.    summaries.append(tf.summary.scalar('learning_rate', lr))    # Add histograms for gradients.    for grad, var in grads:      if grad is not None:        summaries.append(tf.summary.histogram(var.op.name + '/gradients', grad))    # Apply the gradients to adjust the shared variables.    apply_gradient_op = opt.apply_gradients(grads, global_step=global_step)    # Add histograms for trainable variables.    for var in tf.trainable_variables():      summaries.append(tf.summary.histogram(var.op.name, var))    # Track the moving averages of all trainable variables.    variable_averages = tf.train.ExponentialMovingAverage(        cifar10.MOVING_AVERAGE_DECAY, global_step)    variables_averages_op = variable_averages.apply(tf.trainable_variables())    # Group all updates to into a single train op.    train_op = tf.group(apply_gradient_op, variables_averages_op)    # Create a saver.    saver = tf.train.Saver(tf.global_variables())    # Build the summary operation from the last tower summaries.    summary_op = tf.summary.merge(summaries)    # Build an initialization operation to run below.    init = tf.global_variables_initializer()    # Start running operations on the Graph. allow_soft_placement must be set to    # True to build towers on GPU, as some of the ops do not have GPU    # implementations.    sess = tf.Session(config=tf.ConfigProto(        allow_soft_placement=True,        log_device_placement=FLAGS.log_device_placement))    sess.run(init)    # Start the queue runners.    tf.train.start_queue_runners(sess=sess)    summary_writer = tf.summary.FileWriter(FLAGS.train_dir, sess.graph)    for step in xrange(FLAGS.max_steps):      start_time = time.time()      _, loss_value = sess.run([train_op, loss])      duration = time.time() - start_time      assert not np.isnan(loss_value), 'Model diverged with loss = NaN'      if step % 10 == 0:        num_examples_per_step = FLAGS.batch_size * FLAGS.num_gpus        examples_per_sec = num_examples_per_step / duration        sec_per_batch = duration / FLAGS.num_gpus        format_str = ('%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '                      'sec/batch)')        print (format_str % (datetime.now(), step, loss_value,                             examples_per_sec, sec_per_batch))      if step % 100 == 0:        summary_str = sess.run(summary_op)        summary_writer.add_summary(summary_str, step)      # Save the model checkpoint periodically.      if step % 1000 == 0 or (step + 1) == FLAGS.max_steps:        checkpoint_path = os.path.join(FLAGS.train_dir, 'model.ckpt')        saver.save(sess, checkpoint_path, global_step=step)def main(argv=None):  # pylint: disable=unused-argument  cifar10.maybe_download_and_extract()  if tf.gfile.Exists(FLAGS.train_dir):    tf.gfile.DeleteRecursively(FLAGS.train_dir)  tf.gfile.MakeDirs(FLAGS.train_dir)  train()if __name__ == '__main__':tf.app.run()

python cifar10_multi_gpu_train.py --num_gpus=2

---

参考文献

---

http://stackoverflow.com/questions/36668467/change-default-gpu-in-tensorflow
http://stackoverflow.com/questions/37893755/tensorflow-set-cuda-visible-devices-within-jupyter
（原）tensorflow中使用指定的GPU及GPU显存
Using GPUs
TensorFlow官方文档中文版 » 运作方式 » 使用gpu
tensorflow学习笔记（三十一）：构建多GPU代码
cifar10 tutorial
CIFAR10 多 GPU 版本例程源码分析
tensorflow cifar_10 代码阅读与理解