How much VRAM does MOSS-Image need?
MOSS-Image needs about 5.9 GB at 8-bit, or 3.9 GB as a 4-bit GGUF, at its default output size.
- Type
- Image model
- Parameters
- 5.1B
- Text encoder
- Qwen3.5-9B
- Released
- Oct 2026
- License
- Apache 2.0
Memory at each precision
| Precision | Weights | Peak at 1024 × 1024 |
|---|---|---|
| 16-bit BF16 | 9.5 GB | 10.5 GB |
| 8-bit FP8 · Q8_0 | 4.9 GB | 5.9 GB |
| 5-bit GGUF Q5_K_M | 3.4 GB | 4.4 GB |
| 4-bit GGUF Q4_K_M · NVFP4 | 2.8 GB | 3.9 GB |
The Qwen3.5-9B text encoder (7.8 GB at FP8) runs first and then waits in system RAM.
What runs it
| Machine | Best fit | Per image |
|---|---|---|
| RTX 3060 12 GB | 16-bit | 6.3 min |
| RTX 4060 Ti 16 GB | 16-bit | 3.7 min |
| RTX 3090 | 16-bit | 2.3 min |
| RTX 4090 | 16-bit | 59 s |
| RTX 5090 | 16-bit | 47 s |
| RTX PRO 6000 Blackwell | 16-bit | 20 s |
| M4 Max 128 GB | 16-bit | 10 min |
| M5 Max 128 GB | 16-bit | 3.0 min |
| Ryzen AI Max+ 395 128 GB | 16-bit | 4.6 min |
| DGX Spark 128 GB | 16-bit | 2.6 min |
| M3 Ultra 256 GB | 16-bit | 5.9 min |
| M5 Ultra 512 GB | 16-bit | 1.9 min |
| H100 SXM | 16-bit | 11 s |
| DGX Station GB300 | 16-bit | 4.9 s |
1024 × 1024, 50 steps. Times are rough, plus or minus 30%.
Where to get it
Questions
How much VRAM does MOSS-Image need?
MOSS-Image needs about 5.9 GB at 8-bit, or 3.9 GB as a 4-bit GGUF, at its default output size.
Can MOSS-Image run on an RTX 4090?
Yes. At 16-bit it takes about 59 s per image.