Skip to content

Qwen3.8 2.4T A95B on 8× H200 NVL: VRAM, speed and fit

Qwen3.8 2.4T A95B at 2-bit runs on the 8× H200 NVL with 256K of context at about 61 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on 8× H200 NVL · 2-bit (Q2_K) · Ollama / LM Studio · Chat

GPU memory
885 GB
of 1,128 GB
Context
256K
auto
Speed
61 tok/s
42 when full
First reply
0.8 s
2K prompt

Where the memory goes

Model
854 GB
Context
23.5 GB
Overhead
6.9 GB
OS
0.0 GB
Free
243 GB
Memory
885 GB of 1,128 GB
Context
256K tokens · auto
Speed
61 tok/s · 42 when full
First reply
0.8 s for a 2K prompt
Running cost
$65/mo at home · $2,325/mo on a cloud 8× MI300X

Validation: Validated

What Qwen3.8 2.4T A95B on 8× H200 NVL was checked against
CheckTargetResultStatus
Fits in GPU memory1,128 GB usable885 GB neededPass
Headroom5% free243 GB free (22%)Pass
Context for chat16K or more256KPass
Writing speed15 tok/s or more61 tok/sPass
First replyunder 3 s, 2K prompt0.8 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
Qwen3.8 2.4T A95B
Parameters
2.4T, 95B active
Precision
2-bit (Q2_K)
Context
256K, 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 21, 2026.

How to run it

llama.cpp
llama-server -hf unsloth/Qwen3.8-2.4T-A95B-GGUF
vLLM
vllm serve Qwen/Qwen3.8-2.4T-A95B

Qwen3.8 2.4T A95B on similar machines

All machines: Qwen3.8 2.4T A95B VRAM requirements. Everything that runs here: what can the 8× H200 NVL run?

Other models that run on 8× H200 NVL

Questions

Can Qwen3.8 2.4T A95B run on 8× H200 NVL?

Qwen3.8 2.4T A95B at 2-bit runs on the 8× H200 NVL with 256K of context at about 61 tokens a second.

Report VR-00QCXSL, calculated Nov 22, 2026 from published specs, not benchmarked.