# GPU Troubleshooting

# CUDA Error: No Kernel Image Available

If you see this error:

CUDA error: no kernel image is available for execution on the device

This means your GPU is too old for the version of PyTorch installed.

# The Problem

• PyTorch 2.0+ requires CUDA compute capability 7.0 or higher
• Older GPUs (like Quadro M1000M with capability 5.0) are not supported
• PyTorch tries to use the GPU and fails

# The Solution

The app now automatically detects incompatible GPUs and switches to CPU mode. Just run normally:

./run.sh

The app will:

1. Detect your GPU
2. Check if it’s compatible
3. Automatically use CPU if needed
4. Print a message like: “GPU detected (capability 5.0) but incompatible with PyTorch - Using CPU mode”

# Manual Override (If Needed)

If you still see GPU errors, force CPU mode manually:

export CUDA_VISIBLE_DEVICES=-1
./run.sh

Or edit run.sh and add at the top:

export CUDA_VISIBLE_DEVICES=-1

# Performance Impact

With incompatible GPU (using CPU):

• Vocal conversion: ~2-3 min
• Transcription: ~3-4 min
• Video rendering: ~1 min
• Total: ~6-8 minutes for a 4-minute song

With compatible GPU (CUDA 7.0+):

• Vocal conversion: ~30 sec
• Transcription: ~30 sec
• Video rendering: ~30 sec
• Total: ~1.5-2 minutes for a 4-minute song

# GPU Compatibility List

Compatible GPUs (CUDA capability >= 7.0):

• NVIDIA GTX 1660 and newer
• NVIDIA RTX 20xx series and newer
• NVIDIA Tesla V100 and newer
• NVIDIA Quadro RTX series

Incompatible GPUs (CUDA capability < 7.0):

• NVIDIA GTX 900 series (Maxwell)
• NVIDIA GTX 10xx series (some models)
• NVIDIA Quadro M series (M1000M, M2000M, etc.)
• NVIDIA Tesla K series

# Check Your GPU

To check your GPU’s compute capability:

source venv/bin/activate
python3 -c "import torch; print(f'GPU: {torch.cuda.get_device_name(0)}'); print(f'Capability: {torch.cuda.get_device_capability(0)}')"

# Alternative: Install Older PyTorch

If you want to use your older GPU, you could install PyTorch 1.13 (last version supporting CUDA 5.0):

pip uninstall torch torchaudio
pip install torch==1.13.1 torchaudio==0.13.1

Warning: This may cause compatibility issues with other packages. Not recommended.

# Recommended Approach

Just use CPU mode - it works fine! The app is designed to work well on CPU. Processing takes a few extra minutes, but the quality is identical.

# Other GPU Issues

# “CUDA out of memory”

If you get OOM errors:

1. Close other GPU applications
2. Process shorter songs
3. Use CPU mode instead: export CUDA_VISIBLE_DEVICES=-1

# “CUDA initialization failed”

Try:

export CUDA_VISIBLE_DEVICES=-1
./run.sh

# Multiple GPUs

To use a specific GPU:

export CUDA_VISIBLE_DEVICES=0  # Use GPU 0
# or
export CUDA_VISIBLE_DEVICES=1  # Use GPU 1

# Verification

After fixing, run the test script:

source venv/bin/activate
python3 test_setup.py

Look for the GPU section - it should show either:

• “✓ GPU is compatible and will be used for acceleration!”
• “⚠ WARNING: GPU capability X.X is incompatible - The app will automatically use CPU mode”

Both are fine! The app will work either way.

# Still Having Issues?

1. Force CPU mode: export CUDA_VISIBLE_DEVICES=-1
2. Run test: python3 test_setup.py
3. Check logs: Look for “Using CPU mode” messages
4. Report bug: If CPU mode still fails, open an issue on GitHub