Torch Not Compiled With CUDA Enabled in ComfyUI
Fix the 'Torch not compiled with CUDA enabled' error in ComfyUI without blindly reinstalling your whole environment.
If ComfyUI says Torch not compiled with CUDA enabled, the core problem is simple:
your current PyTorch build does not support the GPU path you expected to use.
But the reason it happened is often less simple.
In many broken environments, this error appears after:
- a plugin install changed Torch
- an incompatible wheel was installed
- the environment drifted away from the original GPU setup
- CPU-only packages replaced your CUDA-capable runtime
What this error actually means
ComfyUI is trying to use CUDA, but the installed Torch build cannot do it.
That usually happens in one of these situations:
- you installed CPU-only Torch by mistake
- a plugin dependency replaced the correct Torch build
- your current Torch build does not match your CUDA/runtime combination
- you are using the wrong launch path for the environment
Step 1: Confirm whether the environment used to have GPU support
Ask:
- Did CUDA work before?
- Did this error appear after a plugin install or update?
- Are you sure you are launching the correct environment?
If CUDA worked before and broke later, this is often a drift problem, not an initial install problem.
Step 2: Verify the runtime directly
In the same environment ComfyUI is using, run:
python -c "import torch; print(torch.cuda.is_available())"If it prints False, the environment is not currently usable for CUDA-backed ComfyUI.
You should also check which Torch build is actually installed before making more changes.
Step 3: Repair the core runtime before touching more plugins
Do not keep installing random missing packages until this is fixed.
This is a core runtime problem, so fix the core runtime first:
torchtorchvisiontorchaudio- sometimes
xformers
If a plugin caused the downgrade, repairing only the plugin will not fix the environment.
Why this error is expensive
This is one of the highest-cost failures because it often leads to:
- GPU acceleration disappearing
- users trying repeated
pip installcommands - more package drift
- reinstalling when the environment might still be recoverable
That is why this is not just a CUDA issue. It is often an environment maintenance issue.
A better way to avoid this in the future
If your setup keeps breaking after plugin work, the real problem is usually that the environment has no strong recovery workflow.
Wonderful Launcher is designed to be a better fit when:
- you want to preserve existing workflows and models
- plugin installs keep destabilizing Torch
- you want a more recovery-first path than repeating shell commands
Related guides
- ComfyUI Dependency Conflicts
- ComfyUI Startup Failed? How to Diagnose and Recover Faster
- Installed Custom Nodes and Broke ComfyUI? Recover Without Reinstalling
- ComfyUI Common Issues and Fast Fixes
Best next step
If CUDA worked before and broke after plugin or environment changes, start with Wonderful Launcher.
If the machine is still blocked, use expert help instead of continuing random package repairs.
Wonderful Launcher 文档