How much VRAM does Stable Diffusion 3.5 Large need?
Stable Diffusion 3.5 Large needs about 8.8 GB at 8-bit, or 5.5 GB as a 4-bit GGUF, at its default output size.
- Type
- Image model
- Parameters
- 8.1B
- Text encoder
- CLIP ViT-L/14 + OpenCLIP ViT-bigG/14 + T5-XXL
- Released
- Oct 2024
- License
- Stability AI Community License
Memory at each precision
| Precision | Weights | Peak at 1024 × 1024 |
|---|---|---|
| 16-bit BF16 | 15.2 GB | 16.1 GB |
| 8-bit FP8 · Q8_0 | 7.9 GB | 8.8 GB |
| 5-bit GGUF Q5_K_M | 5.4 GB | 6.4 GB |
| 4-bit GGUF Q4_K_M · NVFP4 | 4.6 GB | 5.5 GB |
The CLIP ViT-L/14 + OpenCLIP ViT-bigG/14 + T5-XXL text encoder (5.2 GB at FP8) runs first and then waits in system RAM.
What runs it
| Machine | Best fit | Per image |
|---|---|---|
| RTX 3060 12 GB | 8-bit | 5.1 min |
| RTX 4060 Ti 16 GB | 8-bit | 2.9 min |
| RTX 3090 | 16-bit | 1.8 min |
| RTX 4090 | 16-bit | 48 s |
| RTX 5090 | 16-bit | 38 s |
| RTX PRO 6000 Blackwell | 16-bit | 16 s |
| M4 Max 128 GB | 16-bit | 8.4 min |
| M5 Max 128 GB | 16-bit | 2.5 min |
| Ryzen AI Max+ 395 128 GB | 16-bit | 3.7 min |
| DGX Spark 128 GB | 16-bit | 2.1 min |
| M3 Ultra 256 GB | 16-bit | 4.7 min |
| M5 Ultra 512 GB | 16-bit | 90 s |
| H100 SXM | 16-bit | 8.8 s |
| DGX Station GB300 | 16-bit | 4.1 s |
1024 × 1024, 28 steps. Times are rough, plus or minus 30%.
Where to get it
Questions
How much VRAM does Stable Diffusion 3.5 Large need?
Stable Diffusion 3.5 Large needs about 8.8 GB at 8-bit, or 5.5 GB as a 4-bit GGUF, at its default output size.
Can Stable Diffusion 3.5 Large run on an RTX 4090?
Yes. At 16-bit it takes about 48 s per image.