RTX 4070 CUDA version and specs overview WSL2

CUDA version on an RTX 4070 under WSL2

Troubleshooting GPU access on a Windows + WSL2 setup. Corrected 2026-10-05: the original text said the 4070 is Ampere; it is NVIDIA Ada (compute capability 8.9).

Key facts

• The GPU model does not pick “the” CUDA version. The Windows driver sets the maximum CUDA version supported; the CUDA toolkit you install (or that PyTorch/TensorFlow wheels bundle) must be at or below it.
• The “CUDA Version” shown by nvidia-smi is the highest CUDA the driver supports, not the installed toolkit. nvcc --version shows the installed toolkit (if any).
• In WSL2 install only the Windows NVIDIA driver; do not install a Linux display driver inside WSL. Use the WSL-Ubuntu CUDA toolkit package, and avoid the generic cuda / cuda-drivers metapackages (they try to install a Linux driver). Per NVIDIA’s WSL guide.
• GeForce RTX 4070: Ada architecture, compute capability 8.9 (NVIDIA CUDA GPUs page).

How to check

1. nvidia-smi (driver version, max CUDA, VRAM).
2. nvcc --version (toolkit, only if installed).
3. In Python: torch.version.cuda and torch.cuda.is_available(); TensorFlow: tf.config.list_physical_devices('GPU').

This machine

Example nvidia-smi output from this machine (WSL2, 2025-02):

+-----------------------------------------------------------------------------------------+  
| NVIDIA-SMI 560.27                 Driver Version: 560.70         CUDA Version: 12.6     |  
|-----------------------------------------+------------------------+----------------------+  
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |  
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |  
|                                         |                        |               MIG M. |  
|=========================================+========================+======================|  
|   0  NVIDIA GeForce RTX 4070 ...    On  |   00000000:01:00.0 Off |                  N/A |  
| N/A   39C    P8              2W /   62W |       0MiB /   8188MiB |      0%      Default |  
|                                         |                        |                  N/A |  
+-----------------------------------------+------------------------+----------------------+  
  
+-----------------------------------------------------------------------------------------+  
| Processes:                                                                              |  
|  GPU   GI   CI        PID   Type   Process name                              GPU Memory |  
|        ID   ID                                                               Usage      |  
|=========================================================================================|  
|  No running processes found                                                             |  
+-----------------------------------------------------------------------------------------+  

Reading: driver 560.70 (Windows), driver-supported CUDA 12.6, GPU “NVIDIA GeForce RTX 4070 …”, 8188 MiB VRAM, 62 W power cap shown. The 8 GB VRAM and low power cap suggest a laptop GPU variant (inference from the output, not confirmed; the desktop card has more VRAM).

Sources

• https://docs.nvidia.com/cuda/wsl-user-guide/index.html (fetched 2026-10-05)
• https://developer.nvidia.com/cuda-gpus (fetched 2026-10-05)
• nvidia-smi output pasted by the vault owner, 2025-02.

Open items

• Exact laptop vs desktop variant and driver version are not re-checked as of 2026-10.