Skip to content

DeepSeek OCR VRAM requirements

How much VRAM DeepSeek OCR needs
4-bit weights: 1.9 GBWith 8K context: 3.1 GB8-bit, 8K context: 4.5 GB
Parameters
3.3B
Active per token
1.1B
Longest context
8K
Released
Oct 2025
License
MIT
Strongest at
Reading documents

How much VRAM DeepSeek OCR needs

Memory DeepSeek OCR needs at each precision and context length
PrecisionWeightsWith 8K
16-bit6.2 GB7.4 GB
8-bit3.3 GB4.5 GB
6-bit2.5 GB3.8 GB
5-bit2.2 GB3.4 GB
4-bit1.9 GB3.1 GB
3-bit1.5 GB2.7 GB
2-bit1.2 GB2.4 GB
1.6-bit656 MB1.8 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run DeepSeek OCR

The best way DeepSeek OCR runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit8K141 tok/s0.4 sPDF: DeepSeek OCR on RTX 4060 report
RTX 50608-bit8K218 tok/s0.3 sPDF: DeepSeek OCR on RTX 5060 report
RTX 30708-bit8K218 tok/s0.3 sPDF: DeepSeek OCR on RTX 3070 report
RTX 3080 10 GB8-bit8K333 tok/s0.2 sPDF: DeepSeek OCR on RTX 3080 10 GB report
RTX 3060 12 GB8-bit8K181 tok/s0.5 sPDF: DeepSeek OCR on RTX 3060 12 GB report
Arc B5808-bit8K153 tok/s0.3 sPDF: DeepSeek OCR on Arc B580 report
RTX 40708-bit8K241 tok/s0.2 sPDF: DeepSeek OCR on RTX 4070 report
RTX 4070 Super8-bit8K241 tok/s0.2 sPDF: DeepSeek OCR on RTX 4070 Super report
RTX 50708-bit8K303 tok/s0.2 sPDF: DeepSeek OCR on RTX 5070 report
RTX 4060 Ti 16 GB8-bit8K149 tok/s0.3 sPDF: DeepSeek OCR on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit8K218 tok/s0.3 sPDF: DeepSeek OCR on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit8K181 tok/s0.3 sPDF: DeepSeek OCR on Arc A770 16 GB report
Radeon RX 9070 XT8-bit8K236 tok/s0.2 sPDF: DeepSeek OCR on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit8K303 tok/s0.1 sPDF: DeepSeek OCR on RTX 4070 Ti Super report
RTX 4080 Super8-bit8K325 tok/s0.1 sPDF: DeepSeek OCR on RTX 4080 Super report
RTX 5070 Ti8-bit8K377 tok/s0.1 sPDF: DeepSeek OCR on RTX 5070 Ti report
RTX 50808-bit8K396 tok/s0.1 sPDF: DeepSeek OCR on RTX 5080 report
Radeon RX 7900 XTX8-bit8K314 tok/s0.3 sPDF: DeepSeek OCR on Radeon RX 7900 XTX report
RTX 30908-bit8K389 tok/s0.2 sPDF: DeepSeek OCR on RTX 3090 report
RTX 40908-bit8K410 tok/s0.1 sPDF: DeepSeek OCR on RTX 4090 report
Radeon AI PRO R97008-bit8K236 tok/s0.2 sPDF: DeepSeek OCR on Radeon AI PRO R9700 report
RTX 50908-bit8K593 tok/s0.1 sPDF: DeepSeek OCR on RTX 5090 report
2× RTX 30908-bit8K389 tok/s0.1 sPDF: DeepSeek OCR on 2× RTX 3090 report
2× RTX 40908-bit8K410 tok/s0.1 sPDF: DeepSeek OCR on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit8K498 tok/s0.1 sPDF: DeepSeek OCR on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit8K396 tok/s0.1 sPDF: DeepSeek OCR on RTX 6000 Ada report
L40S8-bit8K367 tok/s0.1 sPDF: DeepSeek OCR on L40S report
2× RTX 50908-bit8K593 tok/s0.1 sPDF: DeepSeek OCR on 2× RTX 5090 report
4× RTX 30908-bit8K389 tok/s0.1 sPDF: DeepSeek OCR on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit8K593 tok/s0.1 sPDF: DeepSeek OCR on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit8K71 tok/s3.8 sPDF: DeepSeek OCR on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit8K71 tok/s3.8 sPDF: DeepSeek OCR on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit8K71 tok/s3.8 sPDF: DeepSeek OCR on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit8K141 tok/s1.9 sPDF: DeepSeek OCR on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit8K141 tok/s1.9 sPDF: DeepSeek OCR on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit8K141 tok/s1.9 sPDF: DeepSeek OCR on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit8K215 tok/s1.0 sPDF: DeepSeek OCR on M4 Max 128 GB report
M5 Max 128 GB8-bit8K232 tok/s0.3 sPDF: DeepSeek OCR on M5 Max 128 GB report
M3 Ultra 96 GB8-bit8K253 tok/s0.5 sPDF: DeepSeek OCR on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit8K253 tok/s0.5 sPDF: DeepSeek OCR on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit8K253 tok/s0.5 sPDF: DeepSeek OCR on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit8K309 tok/s0.2 sPDF: DeepSeek OCR on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit8K111 tok/s0.6 sPDF: DeepSeek OCR on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit8K111 tok/s0.6 sPDF: DeepSeek OCR on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit8K154 tok/s0.2 sPDF: DeepSeek OCR on RTX Spark report
DGX Spark8-bit8K142 tok/s0.2 sPDF: DeepSeek OCR on DGX Spark report
Dual DGX Spark8-bit8K142 tok/s0.1 sPDF: DeepSeek OCR on Dual DGX Spark report
DGX Station GB3008-bit8K1,037 tok/s0.1 sPDF: DeepSeek OCR on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit8K593 tok/s0.1 sPDF: DeepSeek OCR on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit8K593 tok/s0.1 sPDF: DeepSeek OCR on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit8K593 tok/s0.1 sPDF: DeepSeek OCR on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit8K637 tok/s0.1 sPDF: DeepSeek OCR on A100 80 GB SXM report
H100 SXM8-bit8K808 tok/s0.1 sPDF: DeepSeek OCR on H100 SXM report
4× H200 NVL8-bit8K925 tok/s0.1 sPDF: DeepSeek OCR on 4× H200 NVL report
8× H200 NVL8-bit8K925 tok/s0.1 sPDF: DeepSeek OCR on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf sabafallah/DeepSeek-OCR-GGUF
vLLM
vllm serve deepseek-ai/DeepSeek-OCR

Questions

How much VRAM does DeepSeek OCR need?

DeepSeek OCR needs 1.9 GB for its weights at 4-bit, 3.1 GB with an 8K context and 3.1 GB with 8K. At 8-bit, 4.5 GB.

What is the minimum VRAM for DeepSeek OCR?

About 2.7 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 4060.

How big is DeepSeek OCR?

3.3B parameters (1.1B active per token): 6.2 GB of weights at 16-bit, 1.9 GB at 4-bit.

Can DeepSeek OCR run on an RTX 4090?

Yes. At 8-bit it holds 8K of context and writes about 410 tokens a second.

More models