PyTorch cannot detect RTX 5080 (requires CUDA 12.8) – Nightly build fails to install correctly on Windows

windows
1

Hello, I need help with installing PyTorch for my new NVIDIA RTX 5080 on Windows.

According to the PyTorch contributor reply on GitHub, the oldest CUDA version that supports RTX 5080 is CUDA 12.8.
They also said that any PyTorch build based on CUDA 12.4 / 12.5 cannot detect this GPU.

I tried installing PyTorch Nightly using this command:

pip3 install -U --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu124

But it does not work and my environment cannot detect GPU.

Inside Python, running:

import torch
print(“Torch version:”, torch.version)
print(“CUDA available:”, torch.cuda.is_available())
if torch.cuda.is_available():
print(“GPU:”, torch.cuda.get_device_name(0))

gives errors such as:

ModuleNotFoundError: No module named ‘torch’
NameError: name ‘torch’ is not defined
CUDA not available

So PyTorch fails to install properly, and the GPU is not recognized at all.

My Environment

  • GPU: NVIDIA GeForce RTX 5080

  • OS: Windows 11

  • Python: 3.11.8

  • CUDA Toolkit: (I have not installed CUDA yet because 12.8 is not available on NVIDIA website)

  • Driver version: 581.80 (supports CUDA 13)

  • Goal: Install PyTorch Nightly with CUDA 12.8 or newer so that RTX 5080 is detected

Questions

  1. How can I install a correct PyTorch Nightly build that supports CUDA 12.8 or CUDA 13?

  2. Since NVIDIA has not released CUDA 12.8 toolkit yet, how can we properly set up PyTorch for RTX 5080?

  3. Is there a current PyTorch wheel for Windows that supports the new RTX 5080 GPUs?

  4. Do I need to wait for an official PyTorch/CUDA release?

Additional information

Running nvidia-smi shows the GPU properly:

Driver Version: 581.80
CUDA Version: 13.0
GPU: NVIDIA GeForce RTX 5080

But PyTorch cannot detect it.

Thank you very much! I appreciate any help from the community and developers. :folded_hands:

2

That’s expected since you need a build with CUDA 12.8+ as you already mentioned.

To your questions:

  1. Go to our website, select any stable or nightly build, copy the install command, and paste it into your terminal supporting Python.
  2. That’s wrong as CUDA 12.8 was already released in Jan. 2025.
  3. Yes, all our nightly binaries after the CUDA 12.8 bring up (beginning of 2025) support Blackwell. E.g. you can install our latest stable release (2.9.1) with CUDA 12.8 or 13.0.
  4. PyTorch builds already support Blackwell GPUs since our initial nighty builds created at the beginning of this year and since the stable 2.7 release.
1 Like
3

tri

python -m pip install torch==2.7.1 torchvision==0.22.1 torchaudio==2.7.1 --index-url https://download.pytorch.org/whl/cu128