Ubuntu 16.04/CUDA 10.0 安裝筆記

流程

1. 安裝 NVIDIA 顯卡驅動與相關套件
https://www.tensorflow.org/install/gpu#ubuntu_1604_cuda_10

2. 重開機

3. 測試

nvidia-smi
nvcc -V

4. 編譯測試工具，並進行測試

cd /usr/local/cuda-10.0/samples
sudo make
cd bin/x86_64/linux/release
sudo ./deviceQuery
sudo ./bandwidthTest

5. 安裝 tensorflow-gpu

sudo pip3 install tensorflow-gpu

6. 測試 tensorflow 能否抓到 GPU

python3

from tensorflow.python.client import device_lib

def get_available_gpus():
    local_device_protos = device_lib.list_local_devices()
    return [x.name for x in local_device_protos if x.device_type == 'GPU']

get_available_gpus()

7. 若想防止未來誤更新，可使用 apt-mark 將套件 hold 住

sudo apt-mark hold nvidia-418 cuda-10-0 libcudnn7 libcudnn7-dev

---

參考資料

• GPU support  |  TensorFlow
• python – How to get current available GPUs in tensorflow? – Stack Overflow