ComfyUI prints the device it picked when it starts: look for the Device: line in the terminal. If it names your GPU, ComfyUI is using it; if it says cpu, it is not.
Three ways to check
-
Startup log. ComfyUI logs
Device: <device> <GPU name> : <allocator>for CUDA GPUs, andTotal VRAMandTotal RAMfigures, plus the PyTorch version. -
The API. While it runs:
curl http://127.0.0.1:8188/system_statsThe JSON lists each device with its
name,type,vram_totalandvram_free, plus the PyTorch version. -
PyTorch itself. In the same environment:
python -c "import torch; print(torch.cuda.is_available())"torch.cuda.is_available()returns whether CUDA is currently available.
If it is on the CPU
- Check that you did not start with
--cpu, which runs everything on the CPU. - If PyTorch reports no CUDA, fix the install: the Torch not compiled with CUDA enabled fix.
- Watch free VRAM in
/system_statswhile a job runs, and see how much VRAM ComfyUI needs.
For the surrounding install, see the ComfyUI custom node install guide. If you would rather not set the machine up yourself, the ComfyUI template on Aquanode provisions the box, installs ComfyUI and starts it when you deploy it; once the deployment is Ready, the Overview tab shows the ComfyUI URL.
Sources
- ComfyUI model_management.py (startup log lines). github.com
- ComfyUI server.py (/system_stats). github.com
- PyTorch docs: torch.cuda.is_available. docs.pytorch.org