Kimi K2 Instruct on 8× H200 NVL: VRAM, speed and fit
Kimi K2 Instruct at FP8 runs on the 8× H200 NVL with 128K of context at about 62 tokens a second.
Text generation · Chat · Ollama / LM Studio
Runs comfortably · on 8× H200 NVL · FP8 · Ollama / LM Studio · Chat
- GPU memory
- 973 GB
- of 1,128 GB
- Context
- 128K
- auto
- Speed
- 62 tok/s
- Very fast
- First reply
- 0.3 s
- 2K prompt
Where the memory goes
- Model
- 958 GB
- Context
- 8.6 GB
- Overhead
- 6.0 GB
- OS
- 0.0 GB
- Free
- 155 GB
- Memory
- 973 GB of 1,128 GB
- Context
- 128K tokens · auto
- Speed
- 62 tok/s · very fast
- First reply
- 0.3 s for a 2K prompt
- Running cost
- $65/mo at home · $2,325/mo on a cloud 8× MI300X
Validation: Validated
| Check | Target | Result | Status |
|---|---|---|---|
| Fits in GPU memory | 1,128 GB usable | 973 GB needed | Pass |
| Headroom | 5% free | 155 GB free (14%) | Pass |
| Context for chat | 16K or more | 128K | Pass |
| Writing speed | 15 tok/s or more | 62 tok/s | Pass |
| First reply | under 3 s, 2K prompt | 0.3 s | Pass |
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
- Kimi K2 Instruct
- Parameters
- 1T, 41.5B active
- Precision
- FP8
- Context
- 128K, auto
- Chats at once
- 1
- Use
- Chat
- Runs in
- Ollama / LM Studio
- Hardware
- 8× H200 NVL
- GPU can use
- 1,128 GB of 1,128 GB
Running cost
- At home
- $65/mo
- 2960 W at the wall
- Renting instead
- $2,325/mo
- 8× MI300X at $19/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/Kimi-K2-Instruct-GGUF- vLLM
vllm serve moonshotai/Kimi-K2-Instruct- MLX (Mac)
mlx_lm.chat --model mlx-community/Kimi-K2-Instruct-4bit
Kimi K2 Instruct on similar machines
- 8× RTX PRO 6000 BlackwellRuns
- A100 80 GB SXMToo big
- H100 SXMToo big
- 4× H200 NVLRuns
All machines: Kimi K2 Instruct VRAM requirements. Everything that runs here: what can the 8× H200 NVL run?
Other models that run on 8× H200 NVL
Questions
Can Kimi K2 Instruct run on 8× H200 NVL?
Kimi K2 Instruct at FP8 runs on the 8× H200 NVL with 128K of context at about 62 tokens a second.
Report VR-1RY5CUZ, calculated Nov 22, 2026 from published specs, not benchmarked.