Skip to content

Surya OCR 2 VRAM requirements

How much VRAM Surya OCR 2 needs
4-bit weights: 397 MBWith 8K context: 1.1 GBWith 32K context: 1.4 GB8-bit, 8K context: 1.4 GB
Parameters
686M
Longest context
256K
Released
May 2026
License
OpenRAIL
Strongest at
Reading documents

How much VRAM Surya OCR 2 needs

Memory Surya OCR 2 needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit1.3 GB2.0 GB2.3 GB3.4 GB
8-bit695 MB1.4 GB1.7 GB2.8 GB
6-bit537 MB1.2 GB1.5 GB2.6 GB
5-bit465 MB1.2 GB1.5 GB2.6 GB
4-bit397 MB1.1 GB1.4 GB2.5 GB
3-bit320 MB1.0 GB1.3 GB2.4 GB
2-bit245 MB1.0 GB1.2 GB2.4 GB
1.6-bit135 MB870 MB1.1 GB2.3 GB

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

GPUs and Macs that run Surya OCR 2

The best way Surya OCR 2 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit256K215 tok/s0.1 sPDF: Surya OCR 2 on RTX 4060 report
RTX 50608-bit256K322 tok/s0.1 sPDF: Surya OCR 2 on RTX 5060 report
RTX 30708-bit256K322 tok/s0.1 sPDF: Surya OCR 2 on RTX 3070 report
RTX 3080 10 GB8-bit256K470 tok/s0.1 sPDF: Surya OCR 2 on RTX 3080 10 GB report
RTX 3060 12 GB8-bit256K271 tok/s0.1 sPDF: Surya OCR 2 on RTX 3060 12 GB report
Arc B5808-bit256K212 tok/s0.1 sPDF: Surya OCR 2 on Arc B580 report
RTX 40708-bit256K352 tok/s0.1 sPDF: Surya OCR 2 on RTX 4070 report
RTX 4070 Super8-bit256K352 tok/s0.1 sPDF: Surya OCR 2 on RTX 4070 Super report
RTX 50708-bit256K433 tok/s0.1 sPDF: Surya OCR 2 on RTX 5070 report
RTX 4060 Ti 16 GB8-bit256K226 tok/s0.1 sPDF: Surya OCR 2 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit256K322 tok/s0.1 sPDF: Surya OCR 2 on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit256K243 tok/s0.1 sPDF: Surya OCR 2 on Arc A770 16 GB report
Radeon RX 9070 XT8-bit256K318 tok/s0.1 sPDF: Surya OCR 2 on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit256K433 tok/s0.1 sPDF: Surya OCR 2 on RTX 4070 Ti Super report
RTX 4080 Super8-bit256K460 tok/s0.1 sPDF: Surya OCR 2 on RTX 4080 Super report
RTX 5070 Ti8-bit256K523 tok/s0.1 sPDF: Surya OCR 2 on RTX 5070 Ti report
RTX 50808-bit256K545 tok/s0.1 sPDF: Surya OCR 2 on RTX 5080 report
Radeon RX 7900 XTX8-bit256K400 tok/s0.1 sPDF: Surya OCR 2 on Radeon RX 7900 XTX report
RTX 30908-bit256K537 tok/s0.1 sPDF: Surya OCR 2 on RTX 3090 report
RTX 40908-bit256K562 tok/s0.1 sPDF: Surya OCR 2 on RTX 4090 report
Radeon AI PRO R97008-bit256K318 tok/s0.1 sPDF: Surya OCR 2 on Radeon AI PRO R9700 report
RTX 50908-bit256K759 tok/s0.1 sPDF: Surya OCR 2 on RTX 5090 report
2× RTX 30908-bit256K537 tok/s0.1 sPDF: Surya OCR 2 on 2× RTX 3090 report
2× RTX 40908-bit256K562 tok/s0.1 sPDF: Surya OCR 2 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit256K660 tok/s0.1 sPDF: Surya OCR 2 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit256K545 tok/s0.1 sPDF: Surya OCR 2 on RTX 6000 Ada report
L40S8-bit256K511 tok/s0.1 sPDF: Surya OCR 2 on L40S report
2× RTX 50908-bit256K759 tok/s0.1 sPDF: Surya OCR 2 on 2× RTX 5090 report
4× RTX 30908-bit256K537 tok/s0.1 sPDF: Surya OCR 2 on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit256K759 tok/s0.1 sPDF: Surya OCR 2 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit256K105 tok/s0.9 sPDF: Surya OCR 2 on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit256K105 tok/s0.9 sPDF: Surya OCR 2 on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit256K105 tok/s0.9 sPDF: Surya OCR 2 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit256K190 tok/s0.5 sPDF: Surya OCR 2 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit256K190 tok/s0.5 sPDF: Surya OCR 2 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit256K190 tok/s0.5 sPDF: Surya OCR 2 on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit256K267 tok/s0.2 sPDF: Surya OCR 2 on M4 Max 128 GB report
M5 Max 128 GB8-bit256K282 tok/s0.1 sPDF: Surya OCR 2 on M5 Max 128 GB report
M3 Ultra 96 GB8-bit256K301 tok/s0.1 sPDF: Surya OCR 2 on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit256K301 tok/s0.1 sPDF: Surya OCR 2 on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K301 tok/s0.1 sPDF: Surya OCR 2 on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K347 tok/s0.1 sPDF: Surya OCR 2 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit256K165 tok/s0.1 sPDF: Surya OCR 2 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit256K165 tok/s0.1 sPDF: Surya OCR 2 on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit256K233 tok/s0.1 sPDF: Surya OCR 2 on RTX Spark report
DGX Spark8-bit256K216 tok/s0.1 sPDF: Surya OCR 2 on DGX Spark report
Dual DGX Spark8-bit256K216 tok/s0.1 sPDF: Surya OCR 2 on Dual DGX Spark report
DGX Station GB3008-bit256K1,149 tok/s0.1 sPDF: Surya OCR 2 on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit256K759 tok/s0.1 sPDF: Surya OCR 2 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K759 tok/s0.1 sPDF: Surya OCR 2 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K759 tok/s0.1 sPDF: Surya OCR 2 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit256K804 tok/s0.1 sPDF: Surya OCR 2 on A100 80 GB SXM report
H100 SXM8-bit256K962 tok/s0.1 sPDF: Surya OCR 2 on H100 SXM report
4× H200 NVL8-bit256K1,061 tok/s0.1 sPDF: Surya OCR 2 on 4× H200 NVL report
8× H200 NVL8-bit256K1,061 tok/s0.1 sPDF: Surya OCR 2 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 cstr/surya-det-GGUF
vLLM
vllm serve datalab-to/surya-ocr-2

Questions

How much VRAM does Surya OCR 2 need?

Surya OCR 2 needs 397 MB for its weights at 4-bit, 1.1 GB with an 8K context and 1.4 GB with 32K. At 8-bit, 1.4 GB.

What is the minimum VRAM for Surya OCR 2?

About 1.0 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 Surya OCR 2?

686M parameters: 1.3 GB of weights at 16-bit, 397 MB at 4-bit.

Can Surya OCR 2 run on an RTX 4090?

Yes. At 8-bit it holds 256K of context and writes about 562 tokens a second.

More models