MiniMax-M2 needs about 230 GB of GPU memory for FP8 weights. The smallest fitting setup at 32k context is 2× NVIDIA B200. It supports 21 concurrent conversations with SGLang defaults.
GPU requirements by context length
8k context
GPU
GPUs
Concurrent users
NVIDIA B200
2
64
NVIDIA H200 SXM
2
5
NVIDIA H100 SXM
4
50
NVIDIA H100 NVL
4
64
NVIDIA A100 80GB SXM
4
32
NVIDIA RTX PRO 6000 Blackwell Server Edition
4
64
NVIDIA H20
4
64
NVIDIA L40S
8
64
32k context
GPU
GPUs
Concurrent users
NVIDIA B200
2
21
NVIDIA H200 SXM
2
1
NVIDIA H100 SXM
4
12
NVIDIA H100 NVL
4
24
NVIDIA A100 80GB SXM
4
8
NVIDIA RTX PRO 6000 Blackwell Server Edition
4
26
NVIDIA H20
4
26
NVIDIA L40S
8
19
128k context
GPU
GPUs
Concurrent users
NVIDIA B200
2
5
NVIDIA H200 SXM
4
14
NVIDIA H100 SXM
4
3
NVIDIA H100 NVL
4
6
NVIDIA A100 80GB SXM
4
2
NVIDIA RTX PRO 6000 Blackwell Server Edition
4
6
NVIDIA H20
4
6
NVIDIA L40S
8
4
FP8 weights, SGLang defaults, estimates.
Weights by precision
Weight memory
Precision
Weights
Smallest fitting setup at 32k
FP8
about 230 GB
2x NVIDIA B200
INT4 (AWQ)
about 127 GB
1x NVIDIA B200
Model notes
Attention
Grouped-query attention (GQA)
Architecture
Mixture-of-Experts
Layers
62
Hidden size
3,072
Vocabulary
200,064
At 32k context, 28% of this card's usable memory is working as conversation cache.
Frequently asked questions
Will MiniMaxAI/MiniMax-M2 run on a single H100?
No, a single H100 SXM cannot hold MiniMaxAI/MiniMax-M2 at 32k context with FP8 weights. The model requires multiple GPUs.
What is the cheapest GPU setup for MiniMaxAI/MiniMax-M2?
The cheapest fitting setup at 32k context is 2x NVIDIA B200 with FP8 weights and SGLang defaults.
How much VRAM does MiniMaxAI/MiniMax-M2 need at 128k context?
At 128k context on the cheapest fitting setup, MiniMaxAI/MiniMax-M2 uses about 115 GB of VRAM per GPU for FP8 weights and about 8.3 GB per conversation for the KV cache. The total VRAM needed depends on the GPU count and parallelism configuration.
Can I run MiniMaxAI/MiniMax-M2 with INT4 quantization?
Yes, MiniMaxAI/MiniMax-M2 has INT4 weight figures. Its weights take about 127 GB in INT4 versus about 230 GB in FP8. INT4 can make the model fit on fewer GPUs, but check inference quality for your workload.
Want a different context length, GPU, or quantization? The calculator runs the same numbers live, preloaded with this model.