How much VRAM does Ministral 3 8B need?
Ministral 3 8B needs 5.0 GB for its weights at 4-bit, 6.9 GB with an 8K context and 10.1 GB with 32K. At 8-bit, 10.7 GB.
- Parameters
- 8.9B
- Longest context
- 256K
- Released
- Dec 2025
- License
- Apache 2.0
- Strongest at
- Docs
Memory at each precision
| Precision | Weights | With 8K | With 32K | With 128K |
|---|---|---|---|---|
| As released (FP8) | 9.7 GB | 11.6 GB | 14.8 GB | 27.5 GB |
| 8-bit | 8.8 GB | 10.7 GB | 13.9 GB | 26.7 GB |
| 6-bit | 6.8 GB | 8.7 GB | 11.9 GB | 24.6 GB |
| 5-bit | 5.9 GB | 7.8 GB | 11.0 GB | 23.7 GB |
| 4-bit | 5.0 GB | 6.9 GB | 10.1 GB | 22.9 GB |
| 3-bit | 4.1 GB | 5.9 GB | 9.1 GB | 21.9 GB |
| 2-bit | 3.1 GB | 5.0 GB | 8.2 GB | 20.9 GB |
| 1.6-bit | 1.7 GB | 3.6 GB | 6.8 GB | 19.5 GB |
One conversation, 16-bit context cache, runtime overhead included.
What runs it
| Machine | Best fit | Context | Speed | First reply |
|---|---|---|---|---|
| RTX 3060 12 GB | 6-bit | 28K | 35 tok/s | 1.4 s |
| RTX 4060 Ti 16 GB | 8-bit | 40K | 22 tok/s | 0.8 s |
| RTX 3090 | 8-bit | 96K | 68 tok/s | 0.5 s |
| RTX 4090 | 8-bit | 96K | 73 tok/s | 0.2 s |
| RTX 5090 | 8-bit | 152K | 123 tok/s | 0.2 s |
| 2× RTX 3090 | 8-bit | 256K | 68 tok/s | 0.3 s |
| RTX PRO 6000 Blackwell | 8-bit | 256K | 123 tok/s | 0.1 s |
| M4 Max 128 GB | 8-bit | 256K | 36 tok/s | 2.8 s |
| M5 Max 128 GB | 8-bit | 256K | 40 tok/s | 0.8 s |
| Ryzen AI Max+ 395 128 GB | 8-bit | 256K | 17 tok/s | 1.6 s |
| DGX Spark 128 GB | 8-bit | 256K | 21 tok/s | 0.6 s |
| M3 Ultra 256 GB | 8-bit | 256K | 51 tok/s | 1.6 s |
| M5 Ultra 512 GB | 8-bit | 256K | 71 tok/s | 0.5 s |
| H100 SXM | 8-bit | 256K | 208 tok/s | 0.1 s |
| DGX Station GB300 | 8-bit | 256K | 356 tok/s | 0.1 s |
| 8× H200 NVL | 8-bit | 256K | 272 tok/s | 0.1 s |
Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.
How to run it
- Ollama
ollama run ministral-3:8b- llama.cpp
llama-server -hf unsloth/Ministral-3-8B-Instruct-2512-GGUF- vLLM
vllm serve mistralai/Ministral-3-8B-Instruct-2512- MLX (Mac)
mlx_lm.chat --model mlx-community/Ministral-3-8B-Instruct-2512-4bit
Questions
How much VRAM does Ministral 3 8B need?
Ministral 3 8B needs 5.0 GB for its weights at 4-bit, 6.9 GB with an 8K context and 10.1 GB with 32K. At 8-bit, 10.7 GB.
Can Ministral 3 8B run on an RTX 4090?
Yes. At 8-bit it holds 96K of context and writes about 73 tokens a second.