How much VRAM does MiniMax H3 need?
MiniMax H3 needs about 22.8 GB at 8-bit, or 14.6 GB as a 4-bit GGUF, at its default output size.
- Type
- Video model
- Parameters
- 20.1B
- Text encoder
- Qwen3-VL-32B-Instruct
- Released
- Jul 2026
- License
- MiniMax H3 Community License
Memory at each precision
| Precision | Weights | Peak at 768p, 5 s |
|---|---|---|
| 16-bit BF16 | 37.5 GB | 40.9 GB |
| 8-bit FP8 · Q8_0 | 19.4 GB | 22.8 GB |
| 5-bit GGUF Q5_K_M | 13.3 GB | 16.7 GB |
| 4-bit GGUF Q4_K_M · NVFP4 | 11.2 GB | 14.6 GB |
The Qwen3-VL-32B-Instruct text encoder (31.1 GB at FP8) runs first and then waits in system RAM.
What runs it
| Machine | Best fit | Per clip |
|---|---|---|
| RTX 3060 12 GB | Too big | |
| RTX 4060 Ti 16 GB | Too big | |
| RTX 3090 | Too big | |
| RTX 4090 | Too big | |
| RTX 5090 | 8-bit | 4.9 min |
| RTX PRO 6000 Blackwell | 16-bit | 2.2 min |
| M4 Max 128 GB | 16-bit | 62 min |
| M5 Max 128 GB | 16-bit | 18 min |
| Ryzen AI Max+ 395 128 GB | 16-bit | 28 min |
| DGX Spark 128 GB | 16-bit | 16 min |
| M3 Ultra 256 GB | 16-bit | 35 min |
| M5 Ultra 512 GB | 16-bit | 11 min |
| H100 SXM | 16-bit | 78 s |
| DGX Station GB300 | 16-bit | 43 s |
768p, 5 s, 20 steps. Times are rough, plus or minus 30%.
Where to get it
Questions
How much VRAM does MiniMax H3 need?
MiniMax H3 needs about 22.8 GB at 8-bit, or 14.6 GB as a 4-bit GGUF, at its default output size.
Can MiniMax H3 run on an RTX 4090?
Not comfortably. The smallest machine here that does is the RTX 5090.
More models
Data checked Oct 2026.