ComfyUI Node Optimization: Sparse Attention Boosts Speed by 5-20%
Zironic · reddit · 2026-08-23
A developer released custom Sparse Comfy Kitchen backend nodes for ComfyUI, replacing the default Sparge Attention backend. Key improvements include:
- No need to worry about Sparge installation; required kernels are built-in.
- 5-20% speed increase for the attention part of compute.
- 500MB less VRAM usage.
- Slightly lower quantization error.
Caveats:
- Tested only against comfy prunedint8convrot weights.
- Requires ComfyUI v0.33.0 or later.
- Sparse attention is not free speed; it effectively trades off prompt adherence/quality budget. The impact of density settings on quality depends on the diffusion schedule stage (early steps affect action/adherence, later steps affect motion detail).
More from coding & agent
- raylib 6.2 ships with 300+ improvements and 100 new contributors, 20 new API functions — jsuarez · 2026-10-16
- "Vibecoding" is as misleading a label as "AI" and "SI," dev complains — eschadiol · 2026-10-16
- Doc2Harness: New benchmark finds best coding agent passes only 53% across 329 cases — _akhaliq · 2026-10-16
- Perplexity open-sources SPACE, the Firecracker microVM sandbox behind Perplexity Computer — perplexity_ai · 2026-10-16
- Blackstone goes all-in on Factory for autonomous software development — matanSF · 2026-10-16
- Longtime Windsurf user cancels $10/month legacy plan, citing AI subscription fatigue — tristanbob · 2026-10-16