How much VRAM does FLUX.1 [dev] need?
FLUX.1 [dev] needs about 12.5 GB at 8-bit, or 7.7 GB as a 4-bit GGUF, at its default output size.
- Type
- Image model
- Parameters
- 11.9B
- Text encoder
- CLIP ViT-L/14 + T5-XXL
- Released
- Aug 2024
- License
- FLUX.1 [dev] Non-Commercial License
Memory at each precision
| Precision | Weights | Peak at 1024 × 1024 |
|---|---|---|
| 16-bit BF16 | 22.2 GB | 23.2 GB |
| 8-bit FP8 · Q8_0 | 11.5 GB | 12.5 GB |
| 6-bit GGUF Q6_K | 9.1 GB | 10.2 GB |
| 5-bit GGUF Q5_K_M | 7.9 GB | 8.9 GB |
| 4-bit GGUF Q4_K_M · NVFP4 | 6.6 GB | 7.7 GB |
The CLIP ViT-L/14 + T5-XXL text encoder (4.5 GB at FP8) runs first and then waits in system RAM.
What runs it
| Machine | Best fit | Per image |
|---|---|---|
| RTX 3060 12 GB | 6-bit | 82 s |
| RTX 4060 Ti 16 GB | 8-bit | 41 s |
| RTX 3090 | 16-bit | 26 s |
| RTX 4090 | 16-bit | 12 s |
| RTX 5090 | 16-bit | 9.5 s |
| RTX PRO 6000 Blackwell | 16-bit | 4.6 s |
| M4 Max 128 GB | 16-bit | 2.0 min |
| M5 Max 128 GB | 16-bit | 35 s |
| Ryzen AI Max+ 395 128 GB | 16-bit | 53 s |
| DGX Spark 128 GB | 16-bit | 30 s |
| M3 Ultra 256 GB | 16-bit | 66 s |
| M5 Ultra 512 GB | 16-bit | 21 s |
| H100 SXM | 16-bit | 2.8 s |
| DGX Station GB300 | 16-bit | 1.7 s |
1024 × 1024, 20 steps. Times are rough, plus or minus 30%.
Where to get it
Questions
How much VRAM does FLUX.1 [dev] need?
FLUX.1 [dev] needs about 12.5 GB at 8-bit, or 7.7 GB as a 4-bit GGUF, at its default output size.
Can FLUX.1 [dev] run on an RTX 4090?
Yes. At 16-bit it takes about 12 s per image.