Skip to content

MiniMax M2.5 on 4× H200 NVL: VRAM, speed and fit

MiniMax M2.5 at FP8 runs on the 4× H200 NVL with 192K of context at about 139 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on 4× H200 NVL · FP8 · Ollama / LM Studio · Chat

GPU memory
264 GB
of 564 GB
Context
192K
auto
Speed
139 tok/s
45 when full
First reply
0.2 s
2K prompt

Where the memory goes

Model
214 GB
Context
46.5 GB
Overhead
3.0 GB
OS
0.0 GB
Free
300 GB
Memory
264 GB of 564 GB
Context
192K tokens · auto
Speed
139 tok/s · 45 when full
First reply
0.2 s for a 2K prompt
Running cost
$33/mo at home · $574/mo on a cloud 8× A40

Validation: Validated

What MiniMax M2.5 on 4× H200 NVL was checked against
CheckTargetResultStatus
Fits in GPU memory564 GB usable264 GB neededPass
Headroom5% free300 GB free (53%)Pass
Context for chat16K or more192KPass
Writing speed15 tok/s or more139 tok/sPass
First replyunder 3 s, 2K prompt0.2 sPass

Targets: a reply you can read along with, a first reply the use case tolerates, and 5% of memory left free. Calculated from published specs, not benchmarked.

The setup

Model
MiniMax M2.5
Parameters
229B, 11.6B active
Precision
FP8
Context
192K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
4× H200 NVL
GPU can use
564 GB of 564 GB

Running cost

At home
$33/mo
1520 W at the wall
Renting instead
$574/mo
8× A40 at $4.72/h

4 h a day, power at $0.18 per kWh, cloud at RunPod on-demand prices as of Nov 22, 2026.

How to run it

llama.cpp
llama-server -hf unsloth/MiniMax-M2.5-GGUF
vLLM
vllm serve MiniMaxAI/MiniMax-M2.5
MLX (Mac)
mlx_lm.chat --model mlx-community/MiniMax-M2.5-3bit

MiniMax M2.5 on similar machines

All machines: MiniMax M2.5 VRAM requirements. Everything that runs here: what can the 4× H200 NVL run?

Other models that run on 4× H200 NVL

Questions

Can MiniMax M2.5 run on 4× H200 NVL?

MiniMax M2.5 at FP8 runs on the 4× H200 NVL with 192K of context at about 139 tokens a second.

Report VR-0QQ3WKG, calculated Nov 22, 2026 from published specs, not benchmarked.