Skip to content

DeepSeek V4 Flash on 4× H200 NVL: VRAM, speed and fit

DeepSeek V4 Flash at FP4 + FP8 runs on the 4× H200 NVL with 1M of context at about 122 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on 4× H200 NVL · FP4 + FP8 · Ollama / LM Studio · Chat

GPU memory
164 GB
of 564 GB
Context
1M
auto
Speed
122 tok/s
99 when full
First reply
0.2 s
2K prompt

Where the memory goes

Model
155 GB
Context
6.1 GB
Overhead
2.7 GB
OS
0.0 GB
Free
400 GB
Memory
164 GB of 564 GB
Context
1M tokens · auto
Speed
122 tok/s · 99 when full
First reply
0.2 s for a 2K prompt
Running cost
$33/mo at home · $287/mo on a cloud 4× A40

Validation: Validated

What DeepSeek V4 Flash on 4× H200 NVL was checked against
CheckTargetResultStatus
Fits in GPU memory564 GB usable164 GB neededPass
Headroom5% free400 GB free (71%)Pass
Context for chat16K or more1MPass
Writing speed15 tok/s or more122 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
DeepSeek V4 Flash
Parameters
284B, 13B active
Precision
FP4 + FP8
Context
1M, 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
$287/mo
4× A40 at $2.36/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/DeepSeek-V4-Flash-0731-GGUF
vLLM
vllm serve deepseek-ai/DeepSeek-V4-Flash-0731
MLX (Mac)
mlx_lm.chat --model mlx-community/DeepSeek-V4-Flash-0731-2.4bit-mixed

DeepSeek V4 Flash on similar machines

All machines: DeepSeek V4 Flash VRAM requirements. Everything that runs here: what can the 4× H200 NVL run?

Other models that run on 4× H200 NVL

Questions

Can DeepSeek V4 Flash run on 4× H200 NVL?

DeepSeek V4 Flash at FP4 + FP8 runs on the 4× H200 NVL with 1M of context at about 122 tokens a second.

Report VR-057P8LU, calculated Nov 22, 2026 from published specs, not benchmarked.