Files

T

Frank 689228e60d docs: purge the retired workstation from the knowledge base

Second half of the purge (the first covered the persona, AGENTS.md, README.md,
and the distillation of the per-host pages). Every remaining identifier is
gone: hostname, both LAN addresses, and the Wi-Fi SSID.

34 count-checked substitutions across 22 files — indexes, tool/research pages,
the inventory census, the changelog, one generated HTML report, and five
managed skills. Facts kept, names dropped: a history entry keeps its four-page
listing as a nameless summary, the syslog excerpts keep every timestamp and
line, and the census keeps the .zcompdump rows (the zsh version stamps are the
useful part, the hostname was not).

Deliberately unchanged: install.sh's --host case list and the
packages/<hostname>.pkgs filename, where the hostname is the lookup key.

Left out of scope by Frank's call: agent/memories/** — past-session records the
learn pipeline rewrites.

kb-lint: 0 violations. Residual grep for the hostname, its two IPs, and the
SSID across omp-docs/ and agent/managed-skills/: no hits.

2026-09-28 16:37:38 -07:00

3.1 KiB

Raw Blame History

type, title, description, tags, verified

type

title

description

ComfyUI Troubleshooting

Procedures for diagnosing and fixing ComfyUI on dankclaw.

Service Will Not Start

Check the journal logs first.

ssh dank@dankclaw 'journalctl --user -u comfyui --no-pager -n 100'

"RuntimeError: No HIP GPUs are available"

Cause: The container's ROCm PyTorch cannot see the Strix Halo (gfx1151) GPU. This can happen when /dev/kfd and /dev/dri are not properly passthrough, or when the ROCm PyTorch build doesn't match the GPU architecture.

Fix: Verify GPU passthrough is configured in compose.yaml (should include /dev/kfd and /dev/dri). The container image is comfyui-rocm:gfx1151 which should contain the correct ROCm kernels.

Check that the container image uses the AMD gfx1151 base:

ssh dank@dankclaw 'podman inspect comfyui --format "{{.Config.Image}}"'

If the GPU still doesn't work, ensure the host user is in the render group:

ssh dank@dankclaw 'id dank'

"HIP error: invalid device function"

Cause: The ROCm PyTorch kernels don't match the GPU architecture. The ComfyUI container uses the AMD gfx1151 base image (comfyui-rocm:gfx1151), but if PyTorch was reinstalled without the ROCm build, this error occurs.

Fix: Reinstall the gfx1151 ROCm PyTorch. The container image already ships the correct version (torch 2.9.1+rocm). If the venv was rebuilt outside the container, reinstall:

ssh dank@dankclaw 'cd ~/ComfyUI && source venv/bin/activate && \
  pip uninstall torch torchvision torchaudio pytorch-triton-rocm -y && \
  pip install torch torchvision torchaudio'

Port 8188 Already in Use

ssh dank@dankclaw 'ss -tlnp | grep 8188'

Kill the conflicting process or change the port in the service file.

Model Not Loading

Check that the model file exists in the correct path.

ssh dank@dankclaw 'ls -lh ~/models/checkpoints/*.safetensors'

Verify extra_model_paths.yaml points to the correct base path.

ssh dank@dankclaw 'cat ~/ComfyUI/extra_model_paths.yaml'

Restart ComfyUI after adding or removing models.

ssh dank@dankclaw 'systemctl --user restart comfyui'

Cannot Reach the Web UI

Test connectivity from the workstation.

curl -s -o /dev/null -w "%{http_code}" http://dankclaw:8188

Expected response: 200.

If the response is empty or times out, verify the service is running.

ssh dank@dankclaw 'systemctl --user status comfyui --no-pager'
ssh dank@dankclaw 'ss -tlnp | grep 8188'

Out of Memory

ComfyUI loads the full model into RAM. FLUX.2 klein 4B requires approximately 13 GB. The system has 119 GB total. Close other memory-intensive applications if OOM occurs.

Check memory usage.

ssh dank@dankclaw 'free -h'
ssh dank@dankclaw 'systemctl --user show comfyui --property=MemoryCurrent'

3.1 KiB Raw Blame History