Skip to content

Step 3.5 Flash on Dual DGX Spark: VRAM, speed and fit

Step 3.5 Flash at 8-bit runs on the DGX Spark 128 GB with 72K of context at about 18 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on DGX Spark 128 GB · 8-bit (Q8_0) · Ollama / LM Studio · Chat

GPU memory
241 GB
of 256 GB
Context
72K
auto
Speed
18 tok/s
4.8 when full
First reply
2.5 s
2K prompt

Where the memory goes

Model
197 GB
Context
27.6 GB
Overhead
1.1 GB
OS
15.4 GB
Free
14.7 GB
Memory
241 GB of 256 GB
Context
72K tokens · auto
Speed
18 tok/s · 4.8 when full
First reply
2.5 s for a 2K prompt
Running cost
$3.15/mo at home · $574/mo on a cloud 8× A40

Validation: Validated

What Step 3.5 Flash on Dual DGX Spark was checked against
CheckTargetResultStatus
Fits in GPU memory241 GB usable226 GB neededPass
Headroom5% free14.7 GB free (6%)Pass
Context for chat16K or more72KPass
Writing speed15 tok/s or more18 tok/sPass
First replyunder 3 s, 2K prompt2.5 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
Step 3.5 Flash
Parameters
199B, 9.4B active
Precision
8-bit (Q8_0)
Context
72K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
DGX Spark 128 GB
GPU can use
241 GB of 256 GB

Running cost

At home
$3.15/mo
144 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 ggml-org/Step-3.5-Flash-GGUF
vLLM
vllm serve stepfun-ai/Step-3.5-Flash
MLX (Mac)
mlx_lm.chat --model mlx-community/Step-3.5-Flash-4bit

Step 3.5 Flash on similar machines

All machines: Step 3.5 Flash VRAM requirements. Everything that runs here: what can the Dual DGX Spark run?

Other models that run on Dual DGX Spark

Questions

Can Step 3.5 Flash run on Dual DGX Spark?

Step 3.5 Flash at 8-bit runs on the DGX Spark 128 GB with 72K of context at about 18 tokens a second.

Report VR-109XFR8, calculated Nov 22, 2026 from published specs, not benchmarked.