Tensorflow 2 未在 GPU 上运行

Question

I'm working with tensorflow-gpu version 2.0.0 and I have installed gpu driver and CUDA and cuDNN ( CUDA version 10.1.243_426 and cuDNN v7.6.5.32 and I'm using windows!)

When I compile my model or run:

from tensorflow.python.client import device_lib 
print(device_lib.list_local_devices())

It will print out:

2020-01-12 19:56:50.961755: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2020-01-12 19:56:50.974003: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library nvcuda.dll
2020-01-12 19:56:51.628299: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1618] Found device 0 with properties:
name: GeForce MX150 major: 6 minor: 1 memoryClockRate(GHz): 1.5315
pciBusID: 0000:01:00.0
2020-01-12 19:56:51.636256: I tensorflow/stream_executor/platform/default/dlopen_checker_stub.cc:25] GPU libraries are statically linked, skip dlopen check.
2020-01-12 19:56:51.642106: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1746] Adding visible gpu devices: 0
2020-01-12 19:56:52.386608: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1159] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-01-12 19:56:52.393162: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1165]      0
2020-01-12 19:56:52.396516: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1178] 0:   N
2020-01-12 19:56:52.400632: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1304] Created TensorFlow device (/device:GPU:0 with 1356 MB memory) -> physical GPU (device: 0, na
me: GeForce MX150, pci bus id: 0000:01:00.0, compute capability: 6.1)
[name: "/device:CPU:0"
device_type: "CPU"
memory_limit: 268435456
locality {
}
incarnation: 1008745203605650029
, name: "/device:GPU:0"
device_type: "GPU"
memory_limit: 1422723891
locality {
  bus_id: 1
  links {
  }
}
incarnation: 18036547379173389852
physical_device_desc: "device: 0, name: GeForce MX150, pci bus id: 0000:01:00.0, compute capability: 6.1"
]

Which is saying tensorflow is going to use gpu device for sure! But when I run my model, I can see that gpu isn't doing anything!

However you can see that a part of gpu memory is being used and even I can see a gpu activity which is my program!!

What's going on?! Am I doing something wrong?! I have searched a lot and have checked a lot of questions in SO but nobody asked such a question!

Answer 1

Taken from the official documentation of TensorFlow.

 import tensorflow as tf 
    tf.debugging.set_log_device_placement(True)

    # Create some tensors
    a = tf.constant([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]])
    b = tf.constant([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]])
    c = tf.matmul(a, b)

    print(c)

If you run the code above (which should run on your GPU if your GPU is visible to TensorFlow), then your training will run on TensorFlow.

You must see an output like this one:

Executing op MatMul in device /job:localhost/replica:0/task:0/device:GPU:0 tf.Tensor( [[22. 28.] [49. 64.]], shape=(2, 2), dtype=float32)

Also, you can see that you have a surge in the dedicated GPU memory usage in the Task Manager --> it appears that your GPU is being used, but for certainty run the code above.

Answer 2

Have also noted that Windows Task Manager is not useful for monitoring GPU(dual) activity. Try installing TechPowerUp GPU-Z. (I am running dual NVidia cards). This monitors CPU and GPU activity, power and temperatures.