To check your CUDA version on Windows 10, open Command Prompt and run nvcc --version. The line that says Cuda compilation tools, release 12.x is the CUDA Toolkit version installed on your PC. You can also run nvidia-smi, which shows CUDA Version in the top-right corner, but that number is the highest CUDA version your graphics driver supports, not necessarily the toolkit you have installed.
Two different “CUDA versions”
| Command or tool | What it tells you |
|---|---|
nvcc --version |
The CUDA Toolkit version you installed (used for compiling code) |
nvidia-smi |
Your driver version and the highest CUDA version the driver supports |
| NVIDIA Control Panel > System Information | The CUDA driver version (NVCUDA64.DLL) |
| Python frameworks like PyTorch | The CUDA version the framework was built with |
A driver that supports CUDA 12.8 can run apps built with older toolkits like 12.4 or 11.8. The toolkit can’t be newer than what the driver supports.
Method 1: Check with nvcc
- Press Windows + R, type cmd, and press Enter.
- Type
nvcc --versionand press Enter. - Look for release in the last line, like release 12.4.
If you see ‘nvcc’ is not recognized, the CUDA Toolkit isn’t installed or isn’t in your PATH. Check the next method.
Method 2: Check with nvidia-smi
- In Command Prompt, type
nvidia-smiand press Enter. - Look at the top row for Driver Version and CUDA Version.
nvidia-smi is installed with the NVIDIA driver, so it works even without the CUDA Toolkit.
Method 3: Look in the install folder
Open File Explorer and go to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA. Each installed toolkit has its own folder, like v11.8 or v12.4. You can also check the CUDA_PATH environment variable by running echo %CUDA_PATH% in Command Prompt.
Method 4: NVIDIA Control Panel
- Right-click the desktop and open NVIDIA Control Panel.
- Click Help > System Information > Components.
- Find NVCUDA64.DLL. The product name shows the CUDA driver version.
Check the CUDA version in Python
- PyTorch:
python -c "import torch; print(torch.version.cuda, torch.cuda.is_available())" - TensorFlow:
python -c "import tensorflow as tf; print(tf.sysconfig.get_build_info())"
These show the CUDA version the framework was built for, which matters more than the toolkit version when you install packages.
Frequently asked questions
Why do nvcc and nvidia-smi show different versions?
nvidia-smi shows the maximum version your driver supports. nvcc shows the toolkit you installed. It’s normal for nvidia-smi to show a higher number.
How do I update CUDA?
Update your NVIDIA driver first, then download the CUDA Toolkit version you need from NVIDIA’s developer website. You can install several versions side by side.
Does my GPU support CUDA?
Most NVIDIA GeForce, RTX, and Quadro cards support CUDA. AMD and Intel GPUs don’t. Check NVIDIA’s CUDA GPUs list for your card’s compute capability.
Does this work on Windows 11?
Yes. The commands are the same on Windows 11.
Matthew Burleigh is the head writer at solveyourtech.com, where he covers topics like the iPhone, Microsoft Office, and Google apps. He has a Bachelor’s and Master’s degree in Computer Science and has over 15 years of IT experience.
He has been writing online since 2008 and has published thousands of articles that have been read millions of times.