Skip to content

Inkling Small on 4× H200 NVL: VRAM, speed and fit

Inkling Small at 8-bit runs on the 4× H200 NVL with 4K of context at about 163 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on 4× H200 NVL · 8-bit (Q8_0) · Ollama / LM Studio · Chat

GPU memory
267 GB
of 564 GB
Context
4K
auto
Speed
163 tok/s
Very fast
First reply
0.2 s
2K prompt

Where the memory goes

Model
263 GB
Context
672 MB
Overhead
3.1 GB
OS
0.0 GB
Free
297 GB
Memory
267 GB of 564 GB
Context
4K tokens · auto
Speed
163 tok/s · very fast
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 Inkling Small on 4× H200 NVL was checked against
CheckTargetResultStatus
Fits in GPU memory564 GB usable267 GB neededPass
Headroom5% free297 GB free (53%)Pass
Context for chat4K or more4KPass
Writing speed15 tok/s or more163 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
Inkling Small
Parameters
266B, 11.4B active
Precision
8-bit (Q8_0)
Context
4K, 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 21, 2026.

How to run it

llama.cpp
llama-server -hf unsloth/Inkling-Small-GGUF
vLLM
vllm serve thinkingmachines/Inkling-Small
MLX (Mac)
mlx_lm.chat --model mlx-community/Inkling-Small-mxfp4

Inkling Small on similar machines

All machines: Inkling Small VRAM requirements. Everything that runs here: what can the 4× H200 NVL run?

Other models that run on 4× H200 NVL

Questions

Can Inkling Small run on 4× H200 NVL?

Inkling Small at 8-bit runs on the 4× H200 NVL with 4K of context at about 163 tokens a second.

Report VR-1EQR256, calculated Nov 22, 2026 from published specs, not benchmarked.