How can I tell which version of CUDA is installed?

To determine which version of CUDA is installed on your system, you can use a few different methods, depending on your operating system. Here’s how to check on Windows, Linux, and macOS:

On Windows:

1. Using the NVIDIA System Management Interface (nvidia-smi):

- Open the Command Prompt or PowerShell.

- Type nvidia-smi and press Enter.

- Look for the line that says "CUDA Version". This will display the CUDA version supported by your installed drivers, not necessarily the Toolkit version you have installed.

- Example output:

NVIDIA-SMI 531.79 Driver Version: 531.79 CUDA Version: 12.1

2. Checking the CUDA Toolkit Directory:

- Navigate to the CUDA Toolkit installation directory. The default location is usually C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA.

- Inside, you’ll see a folder named with the version of CUDA installed (e.g., v12.1). The folder name indicates the CUDA version.

3. Using Control Panel

- Search for "Control Panel" and open it.

- Go to "Programs" -> "Programs and Features".

- Look for "NVIDIA CUDA Toolkit" in the list of installed programs. The version number is usually present in the program name.

On Linux:

1. Using the NVIDIA System Management Interface (nvidia-smi):

- Open a terminal.

- Type nvidia-smi and press Enter.

- Find the "CUDA Version" line in the output. This will show the CUDA version supported by your NVIDIA driver.

- Example output:

NVIDIA-SMI 535.104.05 Driver Version: 535.104.05 CUDA Version: 12.2

2. Checking the CUDA Toolkit Directory:

- The CUDA Toolkit is typically installed in /usr/local/cuda or /usr/local/cuda-<version>.

- Check the directory name for the version you have installed (e.g., /usr/local/cuda-12.2).

3. Using nvcc (NVIDIA CUDA Compiler):

- Type nvcc --version or nvcc -V in the terminal and press Enter.

- The output will display the CUDA Toolkit version that the compiler is associated with.

- Example output:

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Tue_May_2_00:41:06_PDT_2023
Cuda compilation tools, release 12.2, V12.2.91
Build cuda_12.2.r12.2/compiler.32831584_0

On macOS:

- CUDA is less common on macOS, as NVIDIA GPUs are not natively supported in recent macOS versions. If you've manually installed it, the methods above for Linux may still apply, specifically checking nvcc --version or the CUDA installation directories.

- You will likely find the Toolkit under /Developer/NVIDIA/CUDA-

Note:

- The CUDA version shown by nvidia-smi refers to the driver's compatibility, while the CUDA Toolkit version found in directories or using nvcc indicates the version you have actually installed.

- It is crucial that the driver version supports the CUDA Toolkit version you are trying to use.

By using these methods, you can reliably check the version of CUDA installed on your system, which is essential for developing and running CUDA-accelerated applications.

Learn more about this