[Bug]: cuda failure #723

Activity

1. digidirt

   added

   bugSomething isn't workingSomething isn't working

   on Sep 29, 2026
2. digidirt commented on Sep 30, 2026

   

   digidirt

   on Sep 30, 2026

   Author

   More actions

   First run latest attempt after clearing registry:

   The backend stopped during startup (exit code: 1).

   Last backend log lines from C:\Program Files\StemDeck-Windows-x64.NVIDIA\data\logs\backend.log:
   return import_from_string(self.app)
   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "C:\Program Files\StemDeck-Windows-x64.NVIDIA\python\Lib\site-packages\uvicorn\importer.py", line 19, in import_from_string
   module = importlib.import_module(module_str)
   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "C:\Program Files\StemDeck-Windows-x64.NVIDIA\python\base\Lib\importlib_init_.py", line 90, in import_module
   return _bootstrap._gcd_import(name[level:], package, level)
   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "", line 1387, in _gcd_import
   File "", line 1360, in _find_and_load
   File "", line 1331, in _find_and_load_unlocked
   File "", line 935, in _load_unlocked
   File "", line 995, in exec_module
   File "", line 488, in call_with_frames_removed
   File "C:\Program Files\StemDeck-Windows-x64.NVIDIA\backend\app\main.py", line 109, in
   "demucs config: model=%s device=%s", DEMUCS_MODEL, get_demucs_device()
   ^^^^^^^^^^^^^^^^^^^
   File "C:\Program Files\StemDeck-Windows-x64.NVIDIA\backend\app\core\settings.py", line 550, in get_demucs_device
   return detect_torch_device() if choice == "auto" else choice
   ^^^^^^^^^^^^^^^^^^^^^
   File "C:\Program Files\StemDeck-Windows-x64.NVIDIA\backend\app\core\config.py", line 59, in detect_torch_device
   return available_torch_devices()[0]
   ^^^^^^^^^^^^^^^^^^^^^^^^^
   File "C:\Program Files\StemDeck-Windows-x64.NVIDIA\backend\app\core\config.py", line 38, in available_torch_devices
   import torch
   File "C:\Program Files\StemDeck-Windows-x64.NVIDIA\python\Lib\site-packages\torch_init.py", line 274, in
   load_dll_libraries()
   File "C:\Program Files\StemDeck-Windows-x64.NVIDIA\python\Lib\site-packages\torch_init.py", line 257, in _load_dll_libraries
   raise err
   OSError: [WinError 127] The specified procedure could not be found. Error loading "C:\Program Files\StemDeck-Windows-x64.NVIDIA\python\Lib\site-packages\torch\lib\c10_cuda.dll" or one of its dependencies.
3. thcp commented on Oct 1, 2026

   

   thcp

   on Oct 1, 2026

   Collaborator

   More actions

   Thanks for the logs, they made this one traceable end to end.

   Your GTX 970 ran into four things in a row:

   1. For a CUDA 13 driver, setup only offered the newest PyTorch build, and that build no longer includes kernels for Maxwell cards like the 970. Older cards are now offered the builds that still support them.
   2. When the CUDA install failed, putting the CPU version back left a CUDA file behind, and that stray file is what stopped the backend from starting. Setup now sets the original files aside first and restores them exactly.
   3. The backend now starts even when that happens, and Settings says clearly that the separation engine needs repairing.
   4. A card that no available build supports is no longer retried on every launch.

   These are in the next release. Since your current install already has the stray file, the new version repairs it on its first start. Let me know how it goes on the 970.
4. thcp

   self-assigned this

   on Oct 1, 2026
5. thcp

   added a commit that references this issue on Oct 1, 2026

   Fix the bugs reported against 0.19.0 (#737)

   ...

   d78fa34
6. thcp

   closed this as completedin #739on Oct 2, 2026
7. meefs

   added a commit that references this issue on Oct 2, 2026

   Release 0.19.1 (stemdeckapp#739)

   ...

   This issue will close once commit 939d66e is merged into the 'main' branch.

   939d66e