PyTorch

How to Fix "NVIDIA GeForce RTX 50xx with CUDA capability sm_120 is not compatible with the current PyTorch installation."

Qualiteg Product Development Team

Sep 10, 2025 — 2 min read

Hello! In this post, we explain how to deal with the following problem in a PyTorch 2.6.0 environment.

NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_70 sm_75 sm_80 sm_86 sm_90.

On other Blackwell GeForce cards, the message looks like this.

NVIDIA GeForce RTX 5080 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_70 sm_75 sm_80 sm_86 sm_90.

NVIDIA GeForce RTX 5070 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_70 sm_75 sm_80 sm_86 sm_90.

NVIDIA GeForce RTX 5060 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_70 sm_75 sm_80 sm_86 sm_90.

NVIDIA GeForce RTX 5060Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_70 sm_75 sm_80 sm_86 sm_90.

This occurs because PyTorch 2.6.0 only supports compute capability levels up to SM_90, so you need to update PyTorch to the latest version.

For details on PyTorch support by GPU (compute capability level), see the following article

●PyTorch Versions and NVIDIA GPU Compute Capability Level Support

https://journal.qualiteg.com/pytorch_and_supported_gpu_version/

Installing the Latest PyTorch (PyTorch 2.8.0)

As of this writing (September 10, 2025), the latest stable release of PyTorch is 2.8.0, so that is the version we will install.

The CUDA release that targets GPUs with SM_120 capability, such as the RTX 5070 Ti, is CUDA 13.0 — but no PyTorch build for CUDA 13.0 has been released yet. So here we install the latest PyTorch v2.8.0 built for CUDA 12.8.

Uninstall the existing PyTorch

pip uninstall torch torchaudio torchvision

Install the latest PyTorch

pip install torch2.8.0+cu128 torchvision0.23.0+cu128 torchaudio==2.8.0+cu128 --index-url https://download.pytorch.org/whl/cu128

With that, everything now runs on the latest GeForce 50 series!

Qualiteg Technology Consulting

Turning the latest GPUs into a production LLM foundation.

Once you have the latest generation like the RTX 5090 up and running, the next question is how to build it into a real foundation. We operate our own GPU clusters and develop LLM products in-house.

From GPU selection to inference optimization (quantization, vLLM) and multi-GPU / multi-node distributed configurations, we offer practical advice grounded in real operational experience.

Explore our LLM infrastructure services →

Reference: Official PyTorch Installation Instructions for the Latest Version

https://pytorch.org/get-started/locally/

Hardening a Raspberry Pi 5: SSH keys, UFW, fail2ban — and the trap where IPv6 comes back after a reboot

A freshly set up Raspberry Pi 5 is wide open: password auth enabled, no firewall, 172 pending updates. We harden it with SSH key-only auth, UFW, fail2ban, and automatic updates, disable IPv6, and verify it all with real reboots — including the trap where IPv6 comes back after a reboot.

Setting Up a Raspberry Pi 5 Without a Monitor: From OS Imaging to SSH Using Only a Windows PC

Set up a Raspberry Pi 5 with no monitor or micro HDMI cable, using only a Windows PC. We use the Raspberry Pi Imager CLI and firstrun.sh to preconfigure Wi-Fi and SSH, with the full procedure and measured timings — about 3 minutes from SD card insertion to an SSH connection.

Developing a LINE Bot Locally: Solving the Ever-Changing Webhook URL Problem with a Fixed URL

Hello! The first wall you hit when building a LINE Bot is not the code — it is the webhook. LINE's Messaging API delivers messages that users send to your bot as HTTPS POST requests from LINE's servers to your server. For this receiving endpoint (the Webhook

Use a Local LLM on Your Home PC from Your Phone While Out: Ollama + Open WebUI + WireCanal

Add a chat interface to the local LLM on your home PC (Ollama + gemma4) with Open WebUI, expose it safely through a tunnel, and use it from your phone while out — a fully measured, step-by-step guide. No Docker, no WSL, and no port forwarding required.