ComfyUI Will Use GPU 0 Only on Windows

If the terminal says Multiple NVIDIA GPUs detected. ComfyUI will use GPU 0 only on Windows by default. To restore all GPUs, pass --cuda-device all --disable-pinned-memory., that is a Windows startup notice, not proof that generation is broken.

For comfyui multi gpu, comfyui select gpu, comfyui use gpu 1, comfyui --cuda-device, or comfyui two gpus, the short answer is:

• one normal workflow usually stays on one selected GPU
• the practical way to use two GPUs is two ComfyUI instances on two ports
• --cuda-device is the safest flag for "use GPU 1 instead of GPU 0"

Splitting one KSampler across cards is not something standard workflows do automatically.

Quick answer

Step 1: Check your GPUs

nvidia-smi

The GPU index is in the left column (first card usually 0, second 1). Then confirm PyTorch sees CUDA:

python -c "import torch; print(torch.cuda.device_count()); print(torch.cuda.get_device_name(0))"

If device_count() returns 0, you have a PyTorch/CUDA install problem, not a multi-GPU one — start with GPU Compatibility.

Step 2: Choose one GPU with --cuda-device

For a manual install:

python main.py --cuda-device 1

For the Windows portable package, edit or copy your launch .bat and add the flag after main.py:

.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --cuda-device 1

Per the official argument description, --cuda-device sets the CUDA device for the instance and hides the others from that process. ComfyUI may still print cuda:0 because the chosen GPU becomes the only visible one.

Step 3: Run two instances on two GPUs

Use different ports so the servers do not collide:

python main.py --cuda-device 0 --port 8188
python main.py --cuda-device 1 --port 8189

Then open http://127.0.0.1:8188 and http://127.0.0.1:8189. This is the most reliable multi-GPU pattern: one queue per GPU.

python main.py --cuda-device 0 --port 8188 --user-directory user-gpu0
python main.py --cuda-device 1 --port 8189 --user-directory user-gpu1

Step 4: Understand --default-device

--default-device sets the default device while keeping other devices visible. That is not automatic multi-GPU execution — use it only when a workflow, custom node, or advanced setup needs other devices to remain visible. For most users, --cuda-device is easier to reason about.

Step 5: Do not use multi GPU to hide a VRAM problem

If a workflow fails because one GPU runs out of VRAM, adding a second GPU usually will not make the same model fit. First try: reduce resolution, reduce batch size, close other GPU apps, use a quantized model, or start with --lowvram. See Python Out of Memory in ComfyUI.

Common mistakes

Verify

Start ComfyUI and read the startup log for your GPU name near the device line (Device: cuda:0 NVIDIA GeForce ...). Then watch nvidia-smi while generating. If you run two instances, queue a small workflow in each and confirm both GPUs show activity. If only one GPU is busy, check the browser port, whether both processes are running, whether each has a different --cuda-device, and whether the second process failed to bind its port.

Related Guides

• GPU Compatibility
• Python Out of Memory in ComfyUI
• ComfyUI Reconnecting Error
• ComfyUI Dependency Conflicts

Source References

• ComfyUI command-line arguments source
• ComfyUI system requirements
• PyTorch CUDA package notes