DeepSeek-OCR 3B — VRAM, Speed & Local Setup

作者： Jakub Rusinowski · 最后更新： 2026年9月11日

Model library → DeepSeek-OCR → DeepSeek-OCR 3B

All 3B parameters stay resident — about 2.6 GB at Q4_K_M, so it runs on effectively any modern GPU — while the MoE decoder activates 6 of 64 routed experts plus 2 shared, roughly 570M parameters per token. Token budgets are explicit and chosen per page: Tiny 64 tokens at 512x512, Small 100 at 640x640, Base 256 at 1024x1024, Large 400 at 1280x1280, and a Gundam mode composing several 640x640 views with one 1024x1024 global view. 8,192-token sequence length. MIT licensed.

DeepSeek-OCR 3B needs about 3 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.

Specifications

Parameters3 Billion (570M active)
Context window8,192
ArchitectureDeepEncoder + DeepSeek-3B MoE decoder
ProviderDeepSeek
LicenceMIT
Specified atQ4_K_M
System RAM8 GB
Record updated2026-09-11

Licence

MIT — commercial use permitted. Commercial use permitted. No usage restrictions beyond attribution.

VRAM and Speed by Quantization

Modelled on a reference NVIDIA RTX 4090 (24 GB), with no KV cache (this record has no published architecture). Speed figures are ESTIMATES from the memory-bandwidth roofline described on the methodology page, not benchmarks we ran — rows marked measured come from published or reader-submitted runs. VRAM here includes the KV cache, so it reads higher than the headline figure above, which does not.

QuantWeightsVRAM neededEst. speedFit on 24 GB
Q2_K1.0 GB1.8 GB~328 tok/s (est.)Fits comfortably
Q3_K_M1.3 GB2.1 GB~320 tok/s (est.)Fits comfortably
Q4_K_M1.8 GB2.6 GB~307 tok/s (est.)Fits comfortably
Q5_K_M2.1 GB2.9 GB~300 tok/s (est.)Fits comfortably
Q6_K2.5 GB3.3 GB~292 tok/s (est.)Fits comfortably
Q8_03.2 GB4.0 GB~278 tok/s (est.)Fits comfortably
F166.0 GB6.8 GB~232 tok/s (est.)Fits comfortably

Want the memory numbers alone, at every quantization level and your own context length? Use the DeepSeek-OCR 3B VRAM calculator.

Buy This HardwareIntel Arc B570 10GB — 10 GB VRAM · 150 W board powerDeploy in the Cloud NowRTX 4090 on RunPod — from $0.34/hr · rate checked 2026-07

or compare on Vast.ai from $0.35/hr (typical low · varies)

As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.

Recommended GPU

The cheapest catalogued GPU that runs DeepSeek-OCR 3B is the Intel Arc B570 (10 GB).

联盟营销声明: 本页部分链接为联盟推广链接——如果你通过它们购买，LLM Configurator 可能会获得佣金，而你无需支付任何额外费用。作为亚马逊联盟成员（Amazon Associate），LLM Configurator 会从符合条件的购买中获得收益。
Intel Arc B570 10GB
10 GB VRAM · 150 W board power
2026年价格波动较大——请以当前商品页价格为准。
在亚马逊查看价格

How to Run DeepSeek-OCR 3B

Install Ollama, then run:

ollama run deepseek-ocr

Weights on Hugging Face: deepseek-ai/DeepSeek-OCR.

Best for: document analysis, vision, multimodal, edge devices.

Can I Run DeepSeek-OCR 3B on My GPU?

DeepSeek-OCR 3B — Frequently Asked Questions

How much VRAM does DeepSeek-OCR 3B need?
About 3 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.
Does DeepSeek-OCR 3B run on an RTX 4090 (24 GB)?
Yes. DeepSeek-OCR 3B needs about 3 GB at Q4_K_M, inside a 24 GB card, at an estimated 307 tokens/sec.
How do I run DeepSeek-OCR 3B locally?
Install Ollama and run `ollama run deepseek-ocr`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.

← All DeepSeek-OCR models | VRAM calculator | Check your own hardware