Batch Any mode 8 steps
Batch recipe: CLIPCached conditioning cache
Caches conditioning/latents to disk: from the second seed on, the whole text-encoding/conditioning stage is skipped (measured 950s→96s per clip on a 24GB card, source: pepikir).
How to run it
- 1 Install CLIPCached nodes into the sampling chain
- 2 Fix the prompt; batch over seeds / small CFG changes
- 3 The first seed runs at normal speed; later seeds speed up markedly
Lossless (bit-identical), but only helps repeated runs of the same prompt.
Representative GPUs
864 × 480 · 8s · 32GB RAM / fast-disk; click a GPU for its full matrix.
| GPU | VRAM | Status | Per clip | VRAM peak |
|---|---|---|---|---|
| RTX 4060 8GB | 8GB | Slow | 7 min 12 s | 7G |
| RTX 3060 12GB | 12GB | OK | 6 min 50 s | 11G |
| RTX 4070 12GB | 12GB | OK | 3 min 59 s | 11G |
| RTX 4070 Ti Super 16GB | 16GB | OK | 2 min 54 s | 15G |
| RTX 4080 Super 16GB | 16GB | OK | 2 min 36 s | 15G |
| RTX 4090 24GB | 24GB | OK | 1 min 53 s | 23G |
| RTX 5090 32GB | 32GB | OK | 1 min 22 s | 26G |
| RTX PRO 6000 Blackwell 96GB | 96GB | OK | 1 min 17 s | 47G |
1344×768 · 5s: 48 of 49 listed GPUs workable, from about 6GB VRAM.
FAQ
What's the minimum GPU for this recipe?
About 6GB+ of VRAM for 1344×768 · 5s (48 of 49 listed GPUs workable). Short 736×416 clips run on less.
How does this recipe combine with other acceleration?
Step distillation picks one option; attention backends (Sage/Kitchen/SLA/VSA) and caches stack on top. See the conflicts table for exact rules.