Skip to main content

ComfyUI Wan 2.2 suddenly slow (0–20% GPU) after "Update All" — the attention-backend fix

Also seen as: "Wan slow after ComfyUI update", "GPU utilization 0% ComfyUI", "sageattention not loading after update", "ComfyUI slow after Update All"

Direct answer

How do I fix "Wan slow / low GPU use after a ComfyUI update"?

Wan (or any diffusion model) that was fast suddenly crawls at 0–20% GPU after a ComfyUI "Update All". The usual cause is that the update broke the compiled attention backend — sage/flash-attention was built against the old torch — so ComfyUI silently fell back to slow PyTorch attention. Low GPU utilization on a card that fits the model means it is CPU-bound, not out of VRAM.

Key facts

  • i"Update All" bumped ComfyUI (and often torch), leaving the sageattention/flash-attention wheel pinned to the previous torch — it fails to load and Comfy falls back to slow SDPA attention.
  • iA newer ComfyUI’s dynamic-VRAM offload spilling weights to system RAM, so the GPU stalls waiting on the CPU.
  • iA custom-node update pulled an incompatible torch/xformers.

Next step

  1. 1Roll ComfyUI back to the pre-update commit — Manager keeps a snapshot; git checkout that commit instead of migrating nodes one at a time.
  2. 2To stay updated, reinstall a sageattention/flash-attention wheel that matches the NEW torch, and confirm it loads at startup (Comfy logs the attention backend).
  3. 3Watch GPU utilization while generating — 0–20% on a model that fits = CPU-bound (attention fallback or RAM offload), not a VRAM wall.
  4. 4If offload is spilling, keep the model resident (raise the VRAM budget; drop --lowvram on a card that actually fits).

If that didn't fix it

  • If utilization is high but it is still slow, that is a different issue (bigger model / higher resolution), not the attention fallback.

Not sure this workflow will run before you rent a GPU?

Paste your ComfyUI workflow JSON into the free checker — it flags missing nodes, model requirements, and the VRAM/GPU you actually need. No signup.

Check my workflow freeSee GPU pricing

Other ComfyUI errors

Fix

GPU Memory Exhausted

Fix

Model Not Found

Fix

Missing Python Package

Fix

Disk Space Exhausted

Fix

Gated Model Access

Fix

missing custom nodes (node type not found)

Frequently Asked Questions

What causes "Wan slow / low GPU use after a ComfyUI update" in ComfyUI?

"Update All" bumped ComfyUI (and often torch), leaving the sageattention/flash-attention wheel pinned to the previous torch — it fails to load and Comfy falls back to slow SDPA attention.

How do I fix "Wan slow / low GPU use after a ComfyUI update"?

Roll ComfyUI back to the pre-update commit — Manager keeps a snapshot; git checkout that commit instead of migrating nodes one at a time. To stay updated, reinstall a sageattention/flash-attention wheel that matches the NEW torch, and confirm it loads at startup (Comfy logs the attention backend).