Installing and Test PyTorch C++ API on Ubuntu with GPU enabled

Install PyTorch with Anaconda for python:

Youtube:

https://www.youtube.com/watch?v=GYbNqcS-o1w

1. Install NVIDIA dirver

Check NVIDIA driver is installed

$ nvidia-smi
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.33.01    Driver Version: 440.33.01    CUDA Version: 10.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 1650    On   | 00000000:01:00.0 Off |                  N/A |
| N/A   37C    P8     3W /  N/A |      0MiB /  3911MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
                                                                           
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

2. Make sure you have installed Anaconda viritual enviroment for PyTorch

https://problemsolvingwithpython.com/01-Orientation/01.05-Installing-Anaconda-on-Linux/

Download 64-Bit (x86) Installer (522 MB)

https://www.anaconda.com/products/individual

3. After installed Anaconda start a new Terminal

(base) user@computer:~$ conda install pytorch torchvision cudatoolkit=10.2 -c pytorch
(base) user@computer:~$ conda list pytorch
(base) user@computer:~$ sudo snap install --classic code

4. Check Anaconda installation With PyTorch for Python wilt Cudatoolkit

(base) user@computer:~$ python
Python 3.7.6 (default, Jan  8 2020, 19:59:22) 
[GCC 7.3.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> print(torch.version.cuda)
10.2
>>> print(torch.__version__)
1.5.0
>>> import numpy as np
>>> print(np.__version__)
1.18.1
>>> exit()
(base) user@computer:~$

5. Create a new viritual enviroment cloned from (base)

(base) user@computer:~$ conda create --name with_cudatoolkit --clone base

6. Activate the new cloned conda enviroment

(base) user@computer:~$ conda activate with_cudatoolkit

Prepare for C++ with CUDA and cuDNN in Anaconda virituell enviroment:

7. Install anaconda cudatoolkit

anaconda / packages / cudatoolkit 10.2.89

https://anaconda.org/anaconda/cudatoolkit

(with_cudatoolkit) user@computer:~$ conda install -c anaconda cudatoolkit

8. Install anaconda cudatoolkit-dev

conda-forge / packages / cudatoolkit-dev 10.1.243

https://anaconda.org/conda-forge/cudatoolkit-dev

(with_cudatoolkit) user@computer:~$ conda install -c conda-forge cudatoolkit-dev

9. Now you can check that CUDA compiler is installed

(with_cudatoolkit) user@computer:~$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Sun_Jul_28_19:07:16_PDT_2019
Cuda compilation tools, release 10.1, V10.1.243
(with_cudatoolkit) user@computer:~$

10. Install anaconda cuDNN

anaconda / packages / cudnn 7.6.5

https://anaconda.org/anaconda/cudnn

(with_cudatoolkit) user@computer:~$ conda install -c anaconda cudnn

Testing PyTorch C++ API with GPU and conda enviroment:

11. Check the homepage PyTorch

Using the PyTorch C++ Frontend

https://pytorch.org/tutorials/advanced/cpp_frontend.html

12. Download PyTorch C++ API

Getting stated

https://pytorch.org/get-started/locally/

I choose:

PyTorch Build:		Stable(1.6.0)
Your OS:		Linux
Package:		LibTorch
Language:		C++/Java
CUDA:			10.2


Run this Command:	Download here (Pre-cxx11 ABI):

https://download.pytorch.org/libtorch/cu102/libtorch-shared-with-deps-1.6.0.zip

Command line download option use wget

$ wget https://download.pytorch.org/libtorch/cu102/libtorch-shared-with-deps-1.6.0.zip

13. Make some directories

(with_cudatoolkit) user@computer:~$ mkdir pytorch_cpp
(with_cudatoolkit) user@computer:~$ cd pytorch_cpp
(with_cudatoolkit) user@computer:~/pytorch_cpp/$ mkdir build
(with_cudatoolkit) user@computer:~/pytorch_cpp/$ cd build
(with_cudatoolkit) user@computer:~/pytorch_cpp/build/$

14. unzip the file

libtorch-shared-with-deps-1.6.0.zip

in

(with_cudatoolkit) user@computer:~/pytorch_cpp/$

folder.

15. Make sure you have installed cmake

16. Copy my files from GitHub

CMakeLists.txt
main.cpp

Put them in folder

(with_cudatoolkit) user@computer:~/pytorch_cpp/

Or make test files from text https://pytorch.org/tutorials/advanced/cpp_frontend.html

17. Do the cmake command with respect to absolute path to the PyTorch C++ API

(with_cudatoolkit) olle@computer:~/pytorch_cpp/build$ cmake -DCMAKE_PREFIX_PATH=/home/olle/pytorch_cpp/libtorch-shared-with-deps-1.6.0/libtorch ..

-- The C compiler identification is GNU 7.5.0
-- The CXX compiler identification is GNU 7.5.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Looking for pthread.h
-- Looking for pthread.h - found
-- Looking for pthread_create
-- Looking for pthread_create - not found
-- Looking for pthread_create in pthreads
-- Looking for pthread_create in pthreads - not found
-- Looking for pthread_create in pthread
-- Looking for pthread_create in pthread - found
-- Found Threads: TRUE  
-- Found CUDA: /home/olle/anaconda3/envs/with_cudatoolkit (found version "10.1") 
-- Caffe2: CUDA detected: 10.1
-- Caffe2: CUDA nvcc is: /home/olle/anaconda3/envs/with_cudatoolkit/bin/nvcc
-- Caffe2: CUDA toolkit directory: /home/olle/anaconda3/envs/with_cudatoolkit
-- Caffe2: Header version is: 10.1
-- Found CUDNN: /home/olle/anaconda3/envs/with_cudatoolkit/lib/libcudnn.so  
-- Found cuDNN: v7.6.5  (include: /home/olle/anaconda3/envs/with_cudatoolkit/include, library: /home/olle/anaconda3/envs/with_cudatoolkit/lib/libcudnn.so)
-- Autodetected CUDA architecture(s):  7.5
-- Added CUDA NVCC flags for: -gencode;arch=compute_75,code=sm_75
-- Found Torch: /home/olle/pytorch_cpp/libtorch-shared-with-deps-1.6.0/libtorch/lib/libtorch.so
-- Configuring done
-- Generating done
-- Build files have been written to: /home/olle/pytorch_cpp/build
(with_cudatoolkit) olle@computer:~/pytorch_cpp/build$

18. Do the make command

(with_cudatoolkit) olle@computer:~/pytorch_cpp/build$ make
Scanning dependencies of target main
[ 50%] Building CXX object CMakeFiles/main.dir/main.cpp.o
[100%] Linking CXX executable main
[100%] Built target main
(with_cudatoolkit) olle@computer:~/pytorch_cpp/build$

19. Run the code

(with_cudatoolkit) olle@computer:~/pytorch_cpp/build$ ./main
 0  0
 0  0
[ CPUFloatType{2,2} ]
CUDA is available! 
 0  0
 0  0
[ CUDAFloatType{2,2} ]
(with_cudatoolkit) olle@computer:~/pytorch_cpp/build$