GPU: Not Available

[nanozouave-oss]

Hi,

First, thanks for your work, you did a great job to create a nice easy to install experience.

The tool run great with great results but using my CPU only. In the server tab it show "GPU: Not Available" but I have a RTX2070 installed which is working with any game, LM studio, etc... How can I provide you more logs to debug ?

Thanks in advance

[Ceneka]

Same here with a 3060ti

[deathreaperz]

Same here with a RTX 4070 Laptop

[Ceneka]

Maybe some missing dependency on the host machine? I downgraded Python from 3.13 to 3.11 to test but same thing

[Charlie22911]

Same issue here, Voicebox v1.11, RTX 4080.
I started the server manually in console to see if there may be any useful output:

2026-01-31 01:58:03,537 - __main__ - INFO - ============================================================ 2026-01-31 01:58:03,538 - __main__ - INFO - voicebox-server starting up... 2026-01-31 01:58:03,538 - __main__ - INFO - Python version: 3.12.10 (tags/v3.12.10:0cc8128, Apr 8 2025, 12:21:36) [MSC v.1943 64 bit (AMD64)] 2026-01-31 01:58:03,539 - __main__ - INFO - Executable: C:\Users\charl\AppData\Local\voicebox\voicebox-server.exe 2026-01-31 01:58:03,539 - __main__ - INFO - Arguments: ['C:\\Users\\charl\\AppData\\Local\\voicebox\\voicebox-server.exe'] 2026-01-31 01:58:03,539 - __main__ - INFO - ============================================================ 2026-01-31 01:58:03,539 - __main__ - INFO - Importing argparse... 2026-01-31 01:58:03,541 - __main__ - INFO - Importing uvicorn... 2026-01-31 01:58:03,614 - __main__ - INFO - Standard library imports successful 2026-01-31 01:58:03,615 - __main__ - INFO - Importing backend.config... 2026-01-31 01:58:03,616 - __main__ - INFO - Importing backend.database... 2026-01-31 01:58:03,770 - __main__ - INFO - Importing backend.main (this may take a while due to torch/transformers)... 2026-01-31 01:58:04,881 - __main__ - INFO - Backend imports successful 2026-01-31 01:58:04,882 - __main__ - INFO - Parsed arguments: host=127.0.0.1, port=8000, data_dir=None 2026-01-31 01:58:04,882 - __main__ - INFO - Initializing database... 2026-01-31 01:58:04,901 - __main__ - INFO - Database initialized successfully 2026-01-31 01:58:04,901 - __main__ - INFO - Starting uvicorn server on 127.0.0.1:8000... INFO: Started server process [37812] INFO: Waiting for application startup. voicebox API starting up... Database initialized at data\voicebox.db Backend: PYTORCH GPU available: None (CPU only) Progress manager initialized with event loop HuggingFace cache directory: C:\Users\charl\.cache\huggingface\hub INFO: Application startup complete. INFO: Uvicorn running on http://127.0.0.1:8000 (Press CTRL+C to quit)

EDIT:

Running the backend manually works fine, I'm guessing the issues is with how the environment in voicebox-server is packaged?

(.venv) PS D:\AI Stuff\voicebox-main> python -m uvicorn backend.main:app --host 127.0.0.1 --port 8000 --reload INFO: Will watch for changes in these directories: ['D:\\AI Stuff\\voicebox-main'] INFO: Uvicorn running on http://127.0.0.1:8000 (Press CTRL+C to quit) INFO: Started reloader process [41900] using WatchFiles INFO: Started server process [13372] INFO: Waiting for application startup. voicebox API starting up... Database initialized at data\voicebox.db Backend: PYTORCH GPU available: CUDA (NVIDIA GeForce RTX 4080 Laptop GPU) Progress manager initialized with event loop HuggingFace cache directory: C:\Users\charl\.cache\huggingface\hub INFO: Application startup complete. INFO: 127.0.0.1:58843 - "OPTIONS /health HTTP/1.1" 200 OK

[jamiepine]

I've been at this all day guys, will get a solution. The CUDA binaries are over 2GB and won't fit as a Github's release asset.

Run softprops/action-gh-release@v1
👩‍🏭 Creating new GitHub release for tag ...
Error: File size (2543828017) is greater than 2 GiB

From a CI run today.

This is proving harder than expected, so apologies for the wait. I'll figure something out.

Alternatively if you want to compile from source, CUDA is working so you can build it locally (just as Claude to help) you'll get the mega speedboost for generations.

[Ceneka]

Running the backend manually works fine, I'm guessing the issues is with how the environment in voicebox-server is packaged?

and if you run it with the port of the app:
python -m uvicorn backend.main:app --host 127.0.0.1 --port 17493 --reload
and then open Voicebox, it works just fine 😃

[jamiepine]

and if you run it with the port of the app: python -m uvicorn backend.main:app --host 127.0.0.1 --port 17493 --reload and then open Voicebox, it works just fine 😃

I wish I had thought to tell people that!

[jamiepine]

Update on the progress: #33

[mimmostilo]

regarding support for amd (rocm) instead? is it more complicated?

[zwxhxndxr]

nd if you run it with the port of the app:

Which directory i should run this?

[Ceneka]

Which directory i should run this?

On ./backend

I also installed lasted Nvidia drivers, Python 3.12 and CUDA drivers on the host machine

[Chovanec]

I have rtx 3060 12gb and it shows gpu not available. Any fix?

[therajat08]

@Ceneka where on windows could i find this directory ?

[backendiain]

@jamiepine Thanks for all your efforts!

FWIW, I found updating the PyTorch stuff then relaunching as @Ceneka said did the trick on Linux with RX 7900XTX GPU.

Not really noticing a performance boost oddly. Might be me that's configured something poorly though.

Only hint of anything awry is when it's generating audio in the terminal console I can see:
MIOpen(HIP): Warning [IsEnoughWorkspace] [GetSolutionsFallback WTI] Solver <GemmFwdRest>, workspace required: 84295680, provided ptr: 0 size: 0

The VRAM looks lower than I'd expect, should be about 24GB VRAM but instead:

Might be I need to fire an env var in before booting the app. Had to do stuff like that for Automatic1111 and ComfyUI on older builds.

Again, great piece of software - thanks for all your work!