Skip to content

DeepSeek OCR 2 on RTX 3060 12 GB: VRAM, speed and fit

DeepSeek OCR 2 at 8-bit runs on the RTX 3060 12 GB with 8K of context at about 174 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on RTX 3060 12 GB · 8-bit (Q8_0) · Ollama / LM Studio · Chat

GPU memory
4.6 GB
of 12.0 GB
Context
8K
auto
Speed
174 tok/s
129 when full
First reply
0.5 s
2K prompt

Where the memory goes

Model
3.4 GB
Context
480 MB
Overhead
753 MB
OS
0.0 GB
Free
7.4 GB
Memory
4.6 GB of 12.0 GB
Context
8K tokens · auto
Speed
174 tok/s · 129 when full
First reply
0.5 s for a 2K prompt
Running cost
$3.98/mo at home · $29/mo on a cloud RTX 2000 Ada

Validation: Validated

What DeepSeek OCR 2 on RTX 3060 12 GB was checked against
CheckTargetResultStatus
Fits in GPU memory12.0 GB usable4.6 GB neededPass
Headroom5% free7.4 GB free (62%)Pass
Context for chat8K or more8KPass
Writing speed15 tok/s or more174 tok/sPass
First replyunder 3 s, 2K prompt0.5 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
DeepSeek OCR 2
Parameters
3.4B, 1.2B active
Precision
8-bit (Q8_0)
Context
8K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
RTX 3060 12 GB
GPU can use
12.0 GB of 12.0 GB

Running cost

At home
$3.98/mo
182 W at the wall
Renting instead
$29/mo
RTX 2000 Ada at $0.24/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 sabafallah/DeepSeek-OCR-2-GGUF
vLLM
vllm serve deepseek-ai/DeepSeek-OCR-2

DeepSeek OCR 2 on similar machines

All machines: DeepSeek OCR 2 VRAM requirements. Everything that runs here: what can the RTX 3060 12 GB run?

Other models that run on RTX 3060 12 GB

Questions

Can DeepSeek OCR 2 run on RTX 3060 12 GB?

DeepSeek OCR 2 at 8-bit runs on the RTX 3060 12 GB with 8K of context at about 174 tokens a second.

Report VR-014MXLX, calculated Nov 22, 2026 from published specs, not benchmarked.