Skip to content

Dots.ocr VRAM requirements

How much VRAM Dots.ocr needs
4-bit weights: 1.7 GBWith 8K context: 2.7 GBWith 32K context: 3.4 GB8-bit, 8K context: 4.0 GB
Parameters
3B
Longest context
128K
Released
Jul 2025
License
MIT
Strongest at
Reading documents

How much VRAM Dots.ocr needs

Memory Dots.ocr needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit5.7 GB6.7 GB7.3 GB9.9 GB
8-bit3.0 GB4.0 GB4.7 GB7.3 GB
6-bit2.3 GB3.3 GB4.0 GB6.6 GB
5-bit2.0 GB3.0 GB3.7 GB6.3 GB
4-bit1.7 GB2.7 GB3.4 GB6.0 GB
3-bit1.4 GB2.4 GB3.0 GB5.7 GB
2-bit1.1 GB2.1 GB2.7 GB5.3 GB
1.6-bit598 MB1.6 GB2.2 GB4.9 GB

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

GPUs and Macs that run Dots.ocr

The best way Dots.ocr runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit128K61 tok/s0.4 sPDF: Dots.ocr on RTX 4060 report
RTX 50608-bit128K97 tok/s0.3 sPDF: Dots.ocr on RTX 5060 report
RTX 30708-bit128K97 tok/s0.3 sPDF: Dots.ocr on RTX 3070 report
RTX 3080 10 GB8-bit128K155 tok/s0.2 sPDF: Dots.ocr on RTX 3080 10 GB report
RTX 3060 12 GB8-bit128K79 tok/s0.4 sPDF: Dots.ocr on RTX 3060 12 GB report
Arc B5808-bit128K68 tok/s0.3 sPDF: Dots.ocr on Arc B580 report
RTX 40708-bit128K108 tok/s0.2 sPDF: Dots.ocr on RTX 4070 report
RTX 4070 Super8-bit128K108 tok/s0.2 sPDF: Dots.ocr on RTX 4070 Super report
RTX 50708-bit128K139 tok/s0.2 sPDF: Dots.ocr on RTX 5070 report
RTX 4060 Ti 16 GB8-bit128K64 tok/s0.3 sPDF: Dots.ocr on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit128K97 tok/s0.2 sPDF: Dots.ocr on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit128K81 tok/s0.3 sPDF: Dots.ocr on Arc A770 16 GB report
Radeon RX 9070 XT8-bit128K109 tok/s0.2 sPDF: Dots.ocr on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit128K139 tok/s0.1 sPDF: Dots.ocr on RTX 4070 Ti Super report
RTX 4080 Super8-bit128K151 tok/s0.1 sPDF: Dots.ocr on RTX 4080 Super report
RTX 5070 Ti8-bit128K179 tok/s0.1 sPDF: Dots.ocr on RTX 5070 Ti report
RTX 50808-bit128K190 tok/s0.1 sPDF: Dots.ocr on RTX 5080 report
Radeon RX 7900 XTX8-bit128K152 tok/s0.3 sPDF: Dots.ocr on Radeon RX 7900 XTX report
RTX 30908-bit128K186 tok/s0.2 sPDF: Dots.ocr on RTX 3090 report
RTX 40908-bit128K197 tok/s0.1 sPDF: Dots.ocr on RTX 4090 report
Radeon AI PRO R97008-bit128K109 tok/s0.2 sPDF: Dots.ocr on Radeon AI PRO R9700 report
RTX 50908-bit128K311 tok/s0.1 sPDF: Dots.ocr on RTX 5090 report
2× RTX 30908-bit128K186 tok/s0.1 sPDF: Dots.ocr on 2× RTX 3090 report
2× RTX 40908-bit128K197 tok/s0.1 sPDF: Dots.ocr on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit128K249 tok/s0.1 sPDF: Dots.ocr on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit128K190 tok/s0.1 sPDF: Dots.ocr on RTX 6000 Ada report
L40S8-bit128K173 tok/s0.1 sPDF: Dots.ocr on L40S report
2× RTX 50908-bit128K311 tok/s0.1 sPDF: Dots.ocr on 2× RTX 5090 report
4× RTX 30908-bit128K186 tok/s0.1 sPDF: Dots.ocr on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit128K311 tok/s0.1 sPDF: Dots.ocr on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit128K31 tok/s3.6 sPDF: Dots.ocr on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit128K31 tok/s3.6 sPDF: Dots.ocr on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit128K31 tok/s3.6 sPDF: Dots.ocr on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit128K64 tok/s1.8 sPDF: Dots.ocr on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit128K64 tok/s1.8 sPDF: Dots.ocr on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit128K64 tok/s1.8 sPDF: Dots.ocr on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit128K105 tok/s0.9 sPDF: Dots.ocr on M4 Max 128 GB report
M5 Max 128 GB8-bit128K115 tok/s0.3 sPDF: Dots.ocr on M5 Max 128 GB report
M3 Ultra 96 GB8-bit128K128 tok/s0.5 sPDF: Dots.ocr on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit128K128 tok/s0.5 sPDF: Dots.ocr on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit128K128 tok/s0.5 sPDF: Dots.ocr on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit128K165 tok/s0.2 sPDF: Dots.ocr on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit128K48 tok/s0.5 sPDF: Dots.ocr on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit128K48 tok/s0.5 sPDF: Dots.ocr on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit128K67 tok/s0.2 sPDF: Dots.ocr on RTX Spark report
DGX Spark8-bit128K61 tok/s0.2 sPDF: Dots.ocr on DGX Spark report
Dual DGX Spark8-bit128K61 tok/s0.1 sPDF: Dots.ocr on Dual DGX Spark report
DGX Station GB3008-bit128K695 tok/s0.1 sPDF: Dots.ocr on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit128K311 tok/s0.1 sPDF: Dots.ocr on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit128K311 tok/s0.1 sPDF: Dots.ocr on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit128K311 tok/s0.1 sPDF: Dots.ocr on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit128K341 tok/s0.1 sPDF: Dots.ocr on A100 80 GB SXM report
H100 SXM8-bit128K474 tok/s0.1 sPDF: Dots.ocr on H100 SXM report
4× H200 NVL8-bit128K579 tok/s0.1 sPDF: Dots.ocr on 4× H200 NVL report
8× H200 NVL8-bit128K579 tok/s0.1 sPDF: Dots.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 mradermacher/dots.ocr-GGUF
vLLM
vllm serve dots-studio/dots.ocr

Questions

How much VRAM does Dots.ocr need?

Dots.ocr needs 1.7 GB for its weights at 4-bit, 2.7 GB with an 8K context and 3.4 GB with 32K. At 8-bit, 4.0 GB.

What is the minimum VRAM for Dots.ocr?

About 2.4 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 Dots.ocr?

3B parameters: 5.7 GB of weights at 16-bit, 1.7 GB at 4-bit.

Can Dots.ocr run on an RTX 4090?

Yes. At 8-bit it holds 128K of context and writes about 197 tokens a second.

More models