Skip to content

Qwen3.5 397B A17B on DGX Spark: VRAM, speed and fit

Qwen3.5 397B A17B at 1.6-bit runs on the DGX Spark 128 GB with 256K of context at about 44 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on DGX Spark 128 GB · 1.6-bit (IQ1_S) · Ollama / LM Studio · Chat

GPU memory
93.7 GB
of 128 GB
Context
256K
auto
Speed
44 tok/s
15 when full
First reply
3.6 s
2K prompt

Where the memory goes

Model
77.5 GB
Context
7.7 GB
Overhead
869 MB
OS
7.7 GB
Free
34.3 GB
Memory
93.7 GB of 128 GB
Context
256K tokens · auto
Speed
44 tok/s · 15 when full
First reply
3.6 s for a 2K prompt
Running cost
$3.15/mo at home · $131/mo on a cloud 4× RTX A5000

Validation: Validated with conditions

What Qwen3.5 397B A17B on DGX Spark was checked against
CheckTargetResultStatus
Fits in GPU memory120 GB usable86.0 GB neededPass
Headroom5% free34.3 GB free (29%)Pass
Context for chat16K or more256KPass
Writing speed15 tok/s or more44 tok/sPass
First replyunder 3 s, 2K prompt3.6 sCheck

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
Qwen3.5 397B A17B
Parameters
403B, 17B active
Precision
1.6-bit (IQ1_S)
Context
256K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
DGX Spark 128 GB
GPU can use
120 GB of 128 GB

Running cost

At home
$3.15/mo
144 W at the wall
Renting instead
$131/mo
4× RTX A5000 at $1.08/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/Qwen3.5-397B-A17B-GGUF
vLLM
vllm serve Qwen/Qwen3.5-397B-A17B
MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen3.5-397B-A17B-4bit

Qwen3.5 397B A17B on similar machines

All machines: Qwen3.5 397B A17B VRAM requirements. Everything that runs here: what can the DGX Spark run?

Other models that run on DGX Spark

Questions

Can Qwen3.5 397B A17B run on DGX Spark?

Qwen3.5 397B A17B at 1.6-bit runs on the DGX Spark 128 GB with 256K of context at about 44 tokens a second.

Report VR-1B0EQU3, calculated Nov 22, 2026 from published specs, not benchmarked.