Serving: Apt-get Install does not use GPU

[dylanrandle]

Hi,

I have tensorflow 1.1.0 GPU version installed with CUDA 8.0 and CUDNN 5.1.10. Tensorflow works fine on GPU. My nvidia driver versions are 375.66.

When I use the new (thank you) installation method with apt-get (sudo apt-get install tensorflow-model-server), and I load my models with:
tensorflow_model_server --port=9000 --enable_batching=true --batching_parameters_file="BatchingParametersProto.txt" --model_config_file="ModelProtoConfig.txt"
it does not use my GPU (tesla K80).

Can anyone explain how to resolve this issue? I have had tensorflow-serving working on GPU before by using bazel build -c opt --config=cuda tensorflow_serving/...

Thanks,
Dylan

[kirilg]

For GPU support, you can continue to use bazel build -c opt --config=cuda tensorflow_serving/..., the apt-get version is currently CPU only.

[dylanrandle]

Thanks for the fast reply. @kirilg

[aniketmaurya]

For GPU support, you can continue to use bazel build -c opt --config=cuda tensorflow_serving/..., the apt-get version is currently CPU only.

is apt-install now availble for GPU ?