Installing CUDA in a Conda Environment

I’ve been wanting to learn CUDA recently. Here are my notes on setting up the environment.

1. Create an environment. We’re just using Conda for environment isolation here — no Python needed.

$ conda create -n cuda-dev
$ conda activate cuda-dev

2. Check the CUDA version with nvidia-smi. Note that this version is the maximum CUDA version supported by the driver, not the CUDA runtime version we’ll install later. When installing, make sure the CUDA runtime version ≤ the driver version.

$ nvidia-smi
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.183.01             Driver Version: 535.183.01   CUDA Version: 12.2     |
|-----------------------------------------+----------------------+----------------------+

3. Install a specific CUDA version.

$ conda install cuda -c nvidia/label/cuda-11.8.0

4. Check the nvcc version to confirm the driver supports this CUDA version.

$ nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Wed_Sep_21_10:33:58_PDT_2022
Cuda compilation tools, release 11.8, V11.8.89
Build cuda_11.8.r11.8/compiler.31833905_0

5. The CUDA installed above does not include development headers and other build-time components. Install the following package as well:

conda install cuda-cudart-dev -c nvidia

The installation of cuda-cudart-dev is based on the instructions here.