MacBook pro 15 (mid-2014) with Nvidia GT 750M 2G (GPU compute capacity 3.0)
Fresh install, don't install any updates.
Installl brew and update to the latest.
brew update
brew upgrade
brew install coreutils
brew install swig # This one not sure its function.
https://github.com/vulgo/webdriver.sh
brew tap vulgo/repo
brew install webdriver.sh
webdriver
Download and install.
https://www.nvidia.com/object/mac-driver-archive.html
Download and install.
https://developer.nvidia.com/cuda-toolkit-archive
https://developer.nvidia.com/rdp/form/cudnn-download-survey
Need register on the Nvidia web then download,
then move them to /usr/local/cuda which is the main cuda directory.
tar zxvf ~/Downloads/cudnn-10.0-osx-x64-v7.4.1.5.tar
cd Downloads/
sudo mv -v cuda/lib/libcudnn* /usr/local/cuda/lib
sudo mv -v cuda/include/cudnn.h /usr/local/cuda/include
(not configured, not used for signle GPU but will be useful for multiple GPUs)
Need register on the Nvidia web then download and move to /usr/local/
xz -d nccl_2.4.2-1+cuda10.0_x86_64.txz
tar xvf nccl_2.4.2-1+cuda10.0_x86_64.tar
cd nccl_2.1.15-1+cuda9.1_x86_64
sudo mkdir -p /usr/local/nccl
sudo mv * /usr/local/nccl
sudo mkdir -p /usr/local/include/third_party/nccl
sudo ln -s /usr/local/nccl/include/nccl.h /usr/local/include/third_party/nccl
Download from the apple app store, change its name to Xcode9.2.app then move it to Application.
Accept license and change to xcode 9.2 developer environment.
sudo xcodebuild -license accept # accept the license
sudo xcode-select -s /Application/Xcode9.2.app/
xcodebuild -version
Compile some CUDA samples to check if the GPU is correctly recognized and supported.
cd /Developer/NVIDIA/CUDA-10.0/samples
sudo make -C 1_Utilities/deviceQuery
./bin/x86_64/darwin/release/deviceQuery
From bazel binary, tensorflow 1.13.1 needs bazel 0.21.0 to compile.
chmod +x bazel-0.21.0-installer-darwin-x86_64.sh
./bazel-<version>-installer-darwin-x86_64.sh --user
Use brew will install lastest bazel but it didn't work to compile tensorflow 1.13.1.
(brew install bazel)
Download and install anaconda then make a virtual environment with python 3.6.8
conda update conda
conda update anaconda
conda create -n tensorflow_gpu python=3.6.8
conda activate tensorflow_gpu
Within python 3.6.8 virtual env, do following:
pip install -U pip six numpy wheel setuptools mock # -U is upgrade.
pip install -U keras_applications==1.0.6 --no-deps
pip install -U keras_preprocessing==1.0.5 --no-deps
Reference https://www.tensorflow.org/install/source
git clone https://github.com/tensorflow/tensorflow.git
cd tensorflow
git status
git checkout r1.13
Disable SIP, in recovery mode (command+R before mac starts)
csrutil status
csrutil disable
After all done enable it in recovery mode.
csrutil status
csrutil enable
Remove all __align(sizeof(T))__ from following 3 files:
tensorflow/core/kernels/depthwise_conv_op_gpu.cu.cc
tensorflow/core/kernels/split_lib_gpu.cu.cc
tensorflow/core/kernels/concat_lib_gpu.impl.cu.cc
For example:
extern shared __align(sizeof(T))__ unsigned char smem[];
becomes
extern shared unsigned char smem[];
Comment out
linkopts = [“-lgomp”]
in
tensorflow/third_party/gpus/cuda/BUILD.tpl
Reference: tensorflow/tensorflow#15172
export PATH="$PATH:$HOME/bin" # This is for installing bazel from binary that bazel locates in #HOME/bin
export TMP=/c/tempdir # This is for bazel that it somehow auto configure $TMP to c:/windows/tmp
export CUDA_HOME=/usr/local/cuda
export DYLD_LIBRARY_PATH=/usr/local/cuda/lib:/usr/local/cuda/extras/CUPTI/lib
export LD_LIBRARY_PATH=$DYLD_LIBRARY_PATH
export PATH=$DYLD_LIBRARY_PATH:$PATH
export PATH="/usr/local/cuda/bin":$PATH
export PATH="/usr/local/opt/llvm/bin:$PATH" # This not sure if it is useful.
Inside tensorflow
./configure
No to unnessary options
Yes to CUDA and cuDNN, compute capacity is 3.0
echo $TMP
sudo mkdir -p $TMP # This is a workaround for bazel, bazel does not create any files in that dir during compiling.
It took more than 1 hour to compile, about 90 mins.
bazel build --config=noaws --config=nogcp --config=nohdfs --config=noignite --config=nokafka --config=nonccl --verbose_failures --config=cuda --config=opt --action_env PATH --action_env LD_LIBRARY_PATH --action_env DYLD_LIBRARY_PATH //tensorflow/tools/pip_package:build_pip_package
bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
sudo pip install /tmp/tensorflow_pkg/tensorflow-1.13.1-cp36-cp36m-macosx_10_7_x86_64.whl
mv /tmp/tensorflow_pkg/tensorflow-1.13.1-cp36-cp36m-macosx_10_7_x86_64.whl ~
import tensorflow as tf
# Creates a graph.
a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')
b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')
c = tf.matmul(a, b)
# Creates a session with log_device_placement set to True.
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
Should let mac switch graphics automaticly so more memory for cuda GPU,
"System Preferences" "Energy Saver".
It gives.
2019-04-06 14:20:18.515459: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:959] OS X does not support NUMA - returning NUMA node zero
2019-04-06 14:20:18.515686: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1433] Found device 0 with properties:
name: GeForce GT 750M major: 3 minor: 0 memoryClockRate(GHz): 0.9255
pciBusID: 0000:01:00.0
totalMemory: 2.00GiB freeMemory: 1.82GiB
2019-04-06 14:20:18.515704: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1512] Adding visible gpu devices: 0
2019-04-06 14:20:19.574943: I tensorflow/core/common_runtime/gpu/gpu_device.cc:984] Device interconnect StreamExecutor with strength 1 edge matrix:
2019-04-06 14:20:19.574972: I tensorflow/core/common_runtime/gpu/gpu_device.cc:990] 0
2019-04-06 14:20:19.574978: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1003] 0: N
2019-04-06 14:20:19.575560: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 1587 MB memory) -> physical GPU (device: 0, name: GeForce GT 750M, pci bus id: 0000:01:00.0, compute capability: 3.0)
Device mapping:
/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: GeForce GT 750M, pci bus id: 0000:01:00.0, compute capability: 3.0
2019-04-06 14:20:19.577180: I tensorflow/core/common_runtime/direct_session.cc:317] Device mapping:
/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: GeForce GT 750M, pci bus id: 0000:01:00.0, compute capability: 3.0
Then.
# Runs the op.
print(sess.run(c))
It should give a matrix in the end.
MatMul: (MatMul): /job:localhost/replica:0/task:0/device:GPU:0
2019-04-06 14:11:02.099622: I tensorflow/core/common_runtime/placer.cc:1059] MatMul (MatMul)/job:localhost/replica:0/task:0/device:GPU:0
a: (Const): /job:localhost/replica:0/task:0/device:GPU:0
2019-04-06 14:11:02.099643: I tensorflow/core/common_runtime/placer.cc:1059] a: (Const)/job:localhost/replica:0/task:0/device:GPU:0
b: (Const): /job:localhost/replica:0/task:0/device:GPU:0
2019-04-06 14:11:02.099655: I tensorflow/core/common_runtime/placer.cc:1059] b: (Const)/job:localhost/replica:0/task:0/device:GPU:0
[[22. 28.]
[49. 64.]]
https://gist.github.com/smitshilu/53cf9ff0fd6cdb64cca69a7e2827ed0f
https://gist.github.com/ageitgey/819a51afa4613649bd18
https://gist.github.com/Willian-Zhang/a3bd10da2d8b343875f3862b2a62eb3b