Skip to content

DeepSeek R1 on DGX Spark: VRAM, speed and fit

DeepSeek R1 at 4-bit needs 388 GB; the DGX Spark 128 GB can use 120 GB. The smallest machine we checked that runs it is the H200 NVL.

Text generation · Chat · Ollama / LM Studio

Won’t fit · on DGX Spark 128 GB · 4-bit (Q4_K_M) · Ollama / LM Studio · Chat

GPU memory
395 GB
of 128 GB
Context
8K
auto
Speed
n/a
First reply
n/a

Where the memory goes

Model
386 GB
Context
549 MB
Overhead
733 MB
OS
7.7 GB
Over by
267 GB
Memory
395 GB of 128 GB
Context
8K tokens · auto
Speed
Doesn’t run here
Running cost
Doesn’t run here · $1,162/mo on a cloud 4× MI300X

Validation: Not validated

What DeepSeek R1 on DGX Spark was checked against
CheckTargetResultStatus
Fits in GPU memory120 GB usable388 GB neededFails
Headroom5% freenoneFails
Context for chat16K or more–Not run
Writing speed15 tok/s or more–Not run
First replyunder 3 s, 2K prompt–Not run

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 R1
Parameters
684B, 40.1B active
Precision
4-bit (Q4_K_M)
Context
8K, 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

Worth trying

  • 4× H200 NVL: Runs it with 160K context
  • 8× RTX PRO 6000 Blackwell: Runs it with 160K context

Running cost

At home
n/a
Doesn’t run on this machine.
Renting instead
$1,162/mo
4× MI300X at $9.56/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-R1-GGUF
vLLM
vllm serve deepseek-ai/DeepSeek-R1
MLX (Mac)
mlx_lm.chat --model mlx-community/DeepSeek-R1-4bit

DeepSeek R1 on similar machines

All machines: DeepSeek R1 VRAM requirements. Everything that runs here: what can the DGX Spark run?

Other models that run on DGX Spark

Questions

Can DeepSeek R1 run on DGX Spark?

DeepSeek R1 at 4-bit needs 388 GB; the DGX Spark 128 GB can use 120 GB. The smallest machine we checked that runs it is the H200 NVL.

Shown at the closest setup. Report VR-1THPEO8, calculated Nov 22, 2026 from published specs, not benchmarked.