Until you can use RTX 2060 on Windows 10 (Installing NVIDIA DRIVER, NVIDIA CUDA toolkit, PyTorch)

at first

I want to use GPU when doing machine learning, but in order to use GPU, it is necessary to install dedicated software. Specifically, there were the following two.

It is a memorandum of each installation method.

NVIDIA DRIVER

-Download from the here site. The capacity is about 600MB, so it will take some time.

image.png

――For my environment, I chose these.

--For the installation options, I chose Custom. --NVIDIA GeForce Experience was already installed, so I unchecked it. --I did a clean install.

NVIDIA CUDA toolkit

-Download CUDA Toolkit xx (xx is the version) from here. --If you plan to use PyTorch, please use 10.2 at this time. --I downloaded the local version. It is over 2.5GB and it takes time to download. --This is also a custom installation. --If you do not have Visual Studio installed, you will be asked for confirmation. If you plan to install Visual Studio, it seems better to install it at individual stages.

PyTorch Select your environment from here. image.png At the bottom, the command will appear, so copy it and execute it at the command prompt.

Let's erase windows and put ubuntu ...

Recommended Posts

Until you can use RTX 2060 on Windows 10 (Installing NVIDIA DRIVER, NVIDIA CUDA toolkit, PyTorch)
Until you can use opencv with python
Until you can use the Google Speech API
Until you use PhantomJS with Python on Heroku
Steps to build PyTorch 1.5 for CUDA 10.2 on Windows
Use RTX 3090 with PyTorch
Use pyvenv on Windows
Use Ansible on Windows
Use QuTiP on Windows
Use pip on Windows
Until you create Python Virtualenv on Windows and launch Jupyter
Build an environment on windows10 where you can try MXNet