增加 GPU 负载 Mac M1 Tensorflow

Question

I'm new to tensorflow and using the GPU on my M1 Mac. Running my code, I observed a max GPU load of about 45%. Is there a way to increase this up to about 100%?

I'm using tensorflow in the following conda environment:

name: tensorflow
 
dependencies:
    - python=3.8
    - pip
    - jupyter
    - apple::tensorflow-deps
    - scikit-learn
    - scipy
    - pandas
    - pandas-datareader
    - matplotlib
    - pillow
    - tqdm
    - requests
    - h5py
    - pyyaml
    - flask
    - boto3
    - openssl
    - pip:
        - tensorflow-macos
        - tensorflow-metal
        - bayesian-optimization
        - gym
        - kaggle

And I also get the following warnings from tensorflow:

I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:305] Could not identify NUMA node of platform GPU ID 0, defaulting to 0. Your kernel may not have been built with NUMA support.

I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:271] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 0 MB memory) -> physical PluggableDevice (device: 0, name: METAL, pci bus id: <undefined>)

W tensorflow/core/platform/profile_utils/cpu_utils.cc:128] Failed to get CPU frequency: 0 Hz

I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.

Answer 1

Your GPU appears to be underutilized because it can't perform faster than you instruct it to.

Typically you can increase the GPU load by increasing the batch dimension -- all else being equal.

For instance, if the batch dimension is 1 with utilization at 45%, if you change the batch dimension to 2 it should be closer to 90%.