GPU vs CPU tensor placement in TensorFlow - Interactive Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to create a tensor on the GPU device.

TensorFlow

import tensorflow as tf
with tf.device('[1]'):
    a = tf.constant([1.0, 2.0, 3.0])

Drag options to blanks, or click blank then click option'

A/CPU:0

B/TPU:0

C/GPU:1

D/GPU:0

Attempts:

3 left

💡 Hint

Common Mistakes

Using '/CPU:0' places the tensor on CPU, not GPU.

Using '/GPU:1' when only one GPU is available causes an error.

The code places the tensor a on the first GPU device using tf.device('/GPU:0').

2fill in blank

medium

Complete the code to check if a tensor is placed on GPU.

TensorFlow

import tensorflow as tf
x = tf.constant([1, 2, 3])
print(x.device.endswith('[1]'))

Drag options to blanks, or click blank then click option'

AGPU:0

BTPU:0

CGPU:1

DCPU:0

Attempts:

3 left

💡 Hint

Common Mistakes

Checking for 'CPU:0' when the tensor is on GPU.

Using 'GPU:1' when the tensor is on 'GPU:0'.

The x.device string ends with the device name where the tensor is placed. Checking if it ends with 'GPU:0' confirms if it's on the first GPU.

3fill in blank

hard

Fix the error in the code to place a tensor on CPU explicitly.

TensorFlow

import tensorflow as tf
with tf.device('[1]'):
    b = tf.constant([4, 5, 6])

Drag options to blanks, or click blank then click option'

A/CPU:0

B/TPU:0

C/GPU:0

D/GPU:1

Attempts:

3 left

💡 Hint

Common Mistakes

Using '/GPU:0' places the tensor on GPU, not CPU.

Using '/TPU:0' causes an error if TPU is not available.

To place a tensor explicitly on the CPU, use tf.device('/CPU:0').

4fill in blank

hard

Fill both blanks to create a tensor on GPU and then move it to CPU.

TensorFlow

import tensorflow as tf
with tf.device('[1]'):
    c = tf.constant([7, 8, 9])
with tf.device('[2]'):
    d = tf.identity(c)

Drag options to blanks, or click blank then click option'

A/GPU:0

B/CPU:0

C/GPU:1

D/TPU:0

Attempts:

3 left

💡 Hint

Common Mistakes

Using '/CPU:0' for the first tensor places it on CPU, not GPU.

Using '/GPU:1' when only one GPU is available causes errors.

The tensor c is created on GPU using /GPU:0. Then d is created as a copy on CPU using /CPU:0.

5fill in blank

hard

Fill all three blanks to create a tensor on CPU, then move it to GPU, and finally check its device.

TensorFlow

import tensorflow as tf
with tf.device('[1]'):
    e = tf.constant([10, 11, 12])
with tf.device('[2]'):
    f = tf.identity(e)
print(f.device.endswith('[3]'))

Drag options to blanks, or click blank then click option'

A/CPU:0

B/GPU:0

CGPU:0

DCPU:0

Attempts:

3 left

💡 Hint

Common Mistakes

Using 'CPU:0' instead of '/CPU:0' in device context causes errors.

Checking device with '/GPU:0' instead of 'GPU:0' in string endswith.

The tensor e is created on CPU (/CPU:0), then copied to GPU (/GPU:0) as f. The print statement checks if f is on GPU by checking if its device string ends with 'GPU:0'.