Skip to content

Llama 3.2 11B Vision Instruct VRAM requirements

How much VRAM Llama 3.2 11B Vision Instruct needs
4-bit weights: 6.0 GBWith 8K context: 8.1 GBWith 32K context: 11.8 GB8-bit, 8K context: 12.6 GB
Parameters
10.7B
Longest context
128K
Released
Sep 2024
License
Llama 3.2 Community
Access
Gated on the Hub
Strongest at
Text and images

Hugging Face asks you to accept its terms before it serves the weights. Its licence (Llama 3.2 Community) is not a standard open-source one: check the terms before commercial use.

How much VRAM Llama 3.2 11B Vision Instruct needs

Memory Llama 3.2 11B Vision Instruct needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit19.9 GB21.9 GB25.7 GB40.7 GB
8-bit10.6 GB12.6 GB16.4 GB31.4 GB
6-bit8.1 GB10.2 GB14.0 GB29.0 GB
5-bit7.1 GB9.1 GB12.9 GB27.9 GB
4-bit6.0 GB8.1 GB11.8 GB26.8 GB
3-bit4.9 GB6.9 GB10.7 GB25.7 GB
2-bit3.7 GB5.8 GB9.5 GB24.5 GB
1.6-bit2.0 GB4.1 GB7.9 GB22.9 GB

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

GPUs and Macs that run Llama 3.2 11B Vision Instruct

The best way Llama 3.2 11B Vision Instruct runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Llama 3.2 11B Vision Instruct on RTX 4060 report
RTX 5060Too bigPDF: Llama 3.2 11B Vision Instruct on RTX 5060 report
RTX 3070Too bigPDF: Llama 3.2 11B Vision Instruct on RTX 3070 report
RTX 3080 10 GB4-bit16K78 tok/s0.7 sPDF: Llama 3.2 11B Vision Instruct on RTX 3080 10 GB report
RTX 3060 12 GB5-bit20K34 tok/s1.7 sPDF: Llama 3.2 11B Vision Instruct on RTX 3060 12 GB report
Arc B5805-bit20K29 tok/s1.2 sPDF: Llama 3.2 11B Vision Instruct on Arc B580 report
RTX 40705-bit20K46 tok/s0.7 sPDF: Llama 3.2 11B Vision Instruct on RTX 4070 report
RTX 4070 Super5-bit20K46 tok/s0.6 sPDF: Llama 3.2 11B Vision Instruct on RTX 4070 Super report
RTX 50705-bit20K60 tok/s0.7 sPDF: Llama 3.2 11B Vision Instruct on RTX 5070 report
RTX 4060 Ti 16 GB8-bit24K18 tok/s1.0 sPDF: Llama 3.2 11B Vision Instruct on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit24K28 tok/s0.9 sPDF: Llama 3.2 11B Vision Instruct on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit24K24 tok/s1.0 sPDF: Llama 3.2 11B Vision Instruct on Arc A770 16 GB report
Radeon RX 9070 XT8-bit24K33 tok/s0.6 sPDF: Llama 3.2 11B Vision Instruct on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit24K41 tok/s0.5 sPDF: Llama 3.2 11B Vision Instruct on RTX 4070 Ti Super report
RTX 4080 Super8-bit24K45 tok/s0.4 sPDF: Llama 3.2 11B Vision Instruct on RTX 4080 Super report
RTX 5070 Ti8-bit24K54 tok/s0.5 sPDF: Llama 3.2 11B Vision Instruct on RTX 5070 Ti report
RTX 50808-bit24K58 tok/s0.4 sPDF: Llama 3.2 11B Vision Instruct on RTX 5080 report
Radeon RX 7900 XTX8-bit72K48 tok/s0.9 sPDF: Llama 3.2 11B Vision Instruct on Radeon RX 7900 XTX report
RTX 30908-bit72K57 tok/s0.6 sPDF: Llama 3.2 11B Vision Instruct on RTX 3090 report
RTX 40908-bit72K61 tok/s0.3 sPDF: Llama 3.2 11B Vision Instruct on RTX 4090 report
Radeon AI PRO R97008-bit120K33 tok/s0.6 sPDF: Llama 3.2 11B Vision Instruct on Radeon AI PRO R9700 report
RTX 50908-bit120K102 tok/s0.2 sPDF: Llama 3.2 11B Vision Instruct on RTX 5090 report
2× RTX 30908-bit128K57 tok/s0.3 sPDF: Llama 3.2 11B Vision Instruct on 2× RTX 3090 report
2× RTX 40908-bit128K61 tok/s0.1 sPDF: Llama 3.2 11B Vision Instruct on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit128K79 tok/s0.3 sPDF: Llama 3.2 11B Vision Instruct on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit128K58 tok/s0.2 sPDF: Llama 3.2 11B Vision Instruct on RTX 6000 Ada report
L40S8-bit128K53 tok/s0.4 sPDF: Llama 3.2 11B Vision Instruct on L40S report
2× RTX 50908-bit128K102 tok/s0.1 sPDF: Llama 3.2 11B Vision Instruct on 2× RTX 5090 report
4× RTX 30908-bit128K57 tok/s0.2 sPDF: Llama 3.2 11B Vision Instruct on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit128K102 tok/s0.2 sPDF: Llama 3.2 11B Vision Instruct on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB4-bit20K15 tok/s14 sPDF: Llama 3.2 11B Vision Instruct on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit24K8.9 tok/s14 sPDF: Llama 3.2 11B Vision Instruct on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit56K8.9 tok/s14 sPDF: Llama 3.2 11B Vision Instruct on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit24K19 tok/s6.8 sPDF: Llama 3.2 11B Vision Instruct on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit128K19 tok/s6.8 sPDF: Llama 3.2 11B Vision Instruct on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit128K19 tok/s6.8 sPDF: Llama 3.2 11B Vision Instruct on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit128K34 tok/s3.4 sPDF: Llama 3.2 11B Vision Instruct on M4 Max 128 GB report
M5 Max 128 GB8-bit128K38 tok/s1.0 sPDF: Llama 3.2 11B Vision Instruct on M5 Max 128 GB report
M3 Ultra 96 GB8-bit128K43 tok/s1.9 sPDF: Llama 3.2 11B Vision Instruct on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit128K43 tok/s1.9 sPDF: Llama 3.2 11B Vision Instruct on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit128K43 tok/s1.9 sPDF: Llama 3.2 11B Vision Instruct on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit128K59 tok/s0.6 sPDF: Llama 3.2 11B Vision Instruct on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit128K14 tok/s2.0 sPDF: Llama 3.2 11B Vision Instruct on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit128K14 tok/s2.0 sPDF: Llama 3.2 11B Vision Instruct on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit128K19 tok/s0.7 sPDF: Llama 3.2 11B Vision Instruct on RTX Spark report
DGX Spark8-bit128K17 tok/s0.7 sPDF: Llama 3.2 11B Vision Instruct on DGX Spark report
Dual DGX Spark8-bit128K17 tok/s0.4 sPDF: Llama 3.2 11B Vision Instruct on Dual DGX Spark report
DGX Station GB3008-bit128K297 tok/s0.1 sPDF: Llama 3.2 11B Vision Instruct on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit128K102 tok/s0.1 sPDF: Llama 3.2 11B Vision Instruct on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit128K102 tok/s0.1 sPDF: Llama 3.2 11B Vision Instruct on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit128K102 tok/s0.1 sPDF: Llama 3.2 11B Vision Instruct on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit128K114 tok/s0.4 sPDF: Llama 3.2 11B Vision Instruct on A100 80 GB SXM report
H100 SXM8-bit128K173 tok/s0.1 sPDF: Llama 3.2 11B Vision Instruct on H100 SXM report
4× H200 NVL8-bit128K227 tok/s0.1 sPDF: Llama 3.2 11B Vision Instruct on 4× H200 NVL report
8× H200 NVL8-bit128K227 tok/s0.1 sPDF: Llama 3.2 11B Vision Instruct 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 leafspark/Llama-3.2-11B-Vision-Instruct-GGUF
vLLM
vllm serve meta-llama/Llama-3.2-11B-Vision-Instruct

Questions

How much VRAM does Llama 3.2 11B Vision Instruct need?

Llama 3.2 11B Vision Instruct needs 6.0 GB for its weights at 4-bit, 8.1 GB with an 8K context and 11.8 GB with 32K. At 8-bit, 12.6 GB.

What is the minimum VRAM for Llama 3.2 11B Vision Instruct?

About 6.9 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 3080 10 GB.

How big is Llama 3.2 11B Vision Instruct?

10.7B parameters: 19.9 GB of weights at 16-bit, 6.0 GB at 4-bit.

Can Llama 3.2 11B Vision Instruct run on an RTX 4090?

Yes. At 8-bit it holds 72K of context and writes about 61 tokens a second.

More models