Skip to content

Chandra OCR 2 VRAM requirements

How much VRAM Chandra OCR 2 needs
4-bit weights: 3.0 GBWith 8K context: 4.3 GBWith 32K context: 5.0 GB8-bit, 8K context: 6.5 GB
Parameters
5.3B
Longest context
256K
Released
Mar 2026
License
OpenRAIL
Strongest at
Reading documents

How much VRAM Chandra OCR 2 needs

Memory Chandra OCR 2 needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit9.9 GB11.2 GB11.9 GB14.9 GB
8-bit5.2 GB6.5 GB7.3 GB10.3 GB
6-bit4.0 GB5.3 GB6.1 GB9.1 GB
5-bit3.5 GB4.8 GB5.6 GB8.6 GB
4-bit3.0 GB4.3 GB5.0 GB8.0 GB
3-bit2.4 GB3.7 GB4.5 GB7.5 GB
2-bit1.8 GB3.1 GB3.9 GB6.9 GB
1.6-bit1.0 GB2.3 GB3.1 GB6.1 GB

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

GPUs and Macs that run Chandra OCR 2

The best way Chandra OCR 2 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit40K32 tok/s0.7 sPDF: Chandra OCR 2 on RTX 4060 report
RTX 50608-bit40K52 tok/s0.5 sPDF: Chandra OCR 2 on RTX 5060 report
RTX 30708-bit40K52 tok/s0.5 sPDF: Chandra OCR 2 on RTX 3070 report
RTX 3080 10 GB8-bit96K86 tok/s0.3 sPDF: Chandra OCR 2 on RTX 3080 10 GB report
RTX 3060 12 GB8-bit160K43 tok/s0.8 sPDF: Chandra OCR 2 on RTX 3060 12 GB report
Arc B5808-bit160K37 tok/s0.6 sPDF: Chandra OCR 2 on Arc B580 report
RTX 40708-bit160K59 tok/s0.4 sPDF: Chandra OCR 2 on RTX 4070 report
RTX 4070 Super8-bit160K59 tok/s0.3 sPDF: Chandra OCR 2 on RTX 4070 Super report
RTX 50708-bit160K77 tok/s0.3 sPDF: Chandra OCR 2 on RTX 5070 report
RTX 4060 Ti 16 GB8-bit256K34 tok/s0.5 sPDF: Chandra OCR 2 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit256K52 tok/s0.4 sPDF: Chandra OCR 2 on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit256K45 tok/s0.5 sPDF: Chandra OCR 2 on Arc A770 16 GB report
Radeon RX 9070 XT8-bit256K61 tok/s0.3 sPDF: Chandra OCR 2 on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit256K77 tok/s0.2 sPDF: Chandra OCR 2 on RTX 4070 Ti Super report
RTX 4080 Super8-bit256K83 tok/s0.2 sPDF: Chandra OCR 2 on RTX 4080 Super report
RTX 5070 Ti8-bit256K100 tok/s0.2 sPDF: Chandra OCR 2 on RTX 5070 Ti report
RTX 50808-bit256K106 tok/s0.2 sPDF: Chandra OCR 2 on RTX 5080 report
Radeon RX 7900 XTX8-bit256K87 tok/s0.5 sPDF: Chandra OCR 2 on Radeon RX 7900 XTX report
RTX 30908-bit256K104 tok/s0.3 sPDF: Chandra OCR 2 on RTX 3090 report
RTX 40908-bit256K111 tok/s0.1 sPDF: Chandra OCR 2 on RTX 4090 report
Radeon AI PRO R97008-bit256K61 tok/s0.3 sPDF: Chandra OCR 2 on Radeon AI PRO R9700 report
RTX 50908-bit256K182 tok/s0.1 sPDF: Chandra OCR 2 on RTX 5090 report
2× RTX 30908-bit256K104 tok/s0.2 sPDF: Chandra OCR 2 on 2× RTX 3090 report
2× RTX 40908-bit256K111 tok/s0.1 sPDF: Chandra OCR 2 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit256K143 tok/s0.2 sPDF: Chandra OCR 2 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit256K106 tok/s0.1 sPDF: Chandra OCR 2 on RTX 6000 Ada report
L40S8-bit256K97 tok/s0.2 sPDF: Chandra OCR 2 on L40S report
2× RTX 50908-bit256K182 tok/s0.1 sPDF: Chandra OCR 2 on 2× RTX 5090 report
4× RTX 30908-bit256K104 tok/s0.1 sPDF: Chandra OCR 2 on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit256K182 tok/s0.1 sPDF: Chandra OCR 2 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit128K17 tok/s6.6 sPDF: Chandra OCR 2 on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit256K17 tok/s6.6 sPDF: Chandra OCR 2 on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit256K17 tok/s6.6 sPDF: Chandra OCR 2 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit256K36 tok/s3.3 sPDF: Chandra OCR 2 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit256K36 tok/s3.3 sPDF: Chandra OCR 2 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit256K36 tok/s3.3 sPDF: Chandra OCR 2 on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit256K61 tok/s1.7 sPDF: Chandra OCR 2 on M4 Max 128 GB report
M5 Max 128 GB8-bit256K67 tok/s0.5 sPDF: Chandra OCR 2 on M5 Max 128 GB report
M3 Ultra 96 GB8-bit256K76 tok/s0.9 sPDF: Chandra OCR 2 on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit256K76 tok/s0.9 sPDF: Chandra OCR 2 on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K76 tok/s0.9 sPDF: Chandra OCR 2 on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K102 tok/s0.3 sPDF: Chandra OCR 2 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit256K26 tok/s1.0 sPDF: Chandra OCR 2 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit256K26 tok/s1.0 sPDF: Chandra OCR 2 on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit256K36 tok/s0.3 sPDF: Chandra OCR 2 on RTX Spark report
DGX Spark8-bit256K33 tok/s0.3 sPDF: Chandra OCR 2 on DGX Spark report
Dual DGX Spark8-bit256K33 tok/s0.2 sPDF: Chandra OCR 2 on Dual DGX Spark report
DGX Station GB3008-bit256K476 tok/s0.1 sPDF: Chandra OCR 2 on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit256K182 tok/s0.1 sPDF: Chandra OCR 2 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K182 tok/s0.1 sPDF: Chandra OCR 2 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K182 tok/s0.1 sPDF: Chandra OCR 2 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit256K203 tok/s0.2 sPDF: Chandra OCR 2 on A100 80 GB SXM report
H100 SXM8-bit256K296 tok/s0.1 sPDF: Chandra OCR 2 on H100 SXM report
4× H200 NVL8-bit256K377 tok/s0.1 sPDF: Chandra OCR 2 on 4× H200 NVL report
8× H200 NVL8-bit256K377 tok/s0.1 sPDF: Chandra 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 prithivMLmods/chandra-ocr-2-GGUF
vLLM
vllm serve datalab-to/chandra-ocr-2

Questions

How much VRAM does Chandra OCR 2 need?

Chandra OCR 2 needs 3.0 GB for its weights at 4-bit, 4.3 GB with an 8K context and 5.0 GB with 32K. At 8-bit, 6.5 GB.

What is the minimum VRAM for Chandra OCR 2?

About 3.7 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 Chandra OCR 2?

5.3B parameters: 9.9 GB of weights at 16-bit, 3.0 GB at 4-bit.

Can Chandra OCR 2 run on an RTX 4090?

Yes. At 8-bit it holds 256K of context and writes about 111 tokens a second.

More models