I successfully built PyTorch from source on Windows 11 using CUDA 12.9 for the RTX 5070. Still working through a small runtime issue, but the full build process is documented here on GitHub: [https://github.com/DMorford/TorchinDesiree]

windows
1

=========================================
PYTORCH RTX 5070 BUILD SESSION SUMMARY

Date: July 24, 2025
Machine: Desiree (i9-14900F, 32GB RAM, RTX 5070 12GB)
CUDA Version: 12.9
Build Status: COMPILATION SUCCESSFUL :white_check_mark:

[DMorford/TorchinDesiree: :fire: Build logs, config notes, and experimental attempts for compiling PyTorch with CUDA support on the bleeding-edge RTX 5070 GPU (Desiree). Raw documentation of GPU pain, progress, and persistence in the quest for AI acceleration.]

(GitHub - DMorford/TorchinDesiree: 🔥 Build logs, config notes, and experimental attempts for compiling PyTorch with CUDA support on the bleeding-edge RTX 5070 GPU (Desiree). Raw documentation of GPU pain, progress, and persistence in the quest for AI acceleration.)

=========================================
WHAT WE ACCOMPLISHED

:white_check_mark: MAJOR SUCCESS: PyTorch Build Completed

  • Ninja build finished successfully
  • All CUDA 12.9 extensions compiled
  • RTX 5070 Blackwell architecture (sm_120) properly targeted
  • Build artifacts generated and installed

:white_check_mark: Environment Setup Perfected

  • x64 Native Tools Command Prompt properly configured
  • Python 3.11.9 virtual environment working
  • CUDA 12.9 + RTX 5070 driver compatibility confirmed
  • All build dependencies installed and verified

:white_check_mark: Build Optimization

  • 18 parallel jobs utilized (perfect for i9-14900F)
  • Response files enabled to prevent Windows linking issues
  • Clean build environment with no conflicting artifacts

=========================================
CURRENT ISSUE

:red_question_mark: Import/Extension Linking Issue

  • PyTorch builds successfully but import fails
  • C extensions not properly linked to Python environment
  • Common issue with Windows custom PyTorch builds
  • Error: “Failed to load PyTorch C extensions”
2

The build is now complete and running solid on both RTX 5070 and RTX 5070 Ti with no issues on Windows 11. Just wanted to close the loop in case anyone else is looking for a ready-to-use setup.

Hardware: RTX 5070 12GB & RTX 5070Ti 16GB (Blackwell sm_120)

  • CUDA: 12.8.0 (Working) vs 12.9.0 (Failed)

  • PyTorch: 2.9.0.dev20250803+cu128 (Nightly)

  • AI Assistant: Claude Code by Anthropic

  • OS: Windows 11 Professional

  • IDE: Visual Studio 2022 Community