Meta
ChatReasoningMulti-purpose11B
Chat

Llama 3.2 11B Vision

Llama · Llama 3.2 Community License

Meta's first multimodal Llama model. Handles both text and image inputs. At 11B parameters, fits comfortably on 16GB GPUs at Q4. Vision capabilities are useful for image understanding tasks but text quality is comparable to Llama 3.1 8B, not 70B.

Parameters
11B
Architecture
Dense
Context
131,072 tokens
Released
2024-09-25
Engines
llama.cpp, ollama, vLLM
Builder Tools
LM Studio, Ollama, Open WebUI

Parameters

11B

VRAM

10 GB

Context

128K

Formats

5

GPUs

34

Llama 3.2 11B Vision (11B) requires 10 GB VRAM at recommended quality (Q6_K). At efficient quality (Q4_K_M), it fits in 7.2 GB VRAM, making it compatible with the NVIDIA RTX 4090 Laptop (150-175W). On NVIDIA Grace Blackwell Ultra GB300, expect approximately 260 tok/s at Q8_0. For the best experience, Starter AI Desktop ($582) is recommended.

Source: OwnRig methodology

VRAM (Recommended)

10 GB

Quantization

Q6_K

File Size

8.8 GB

Max Context

128K tokens

Primary Use

Chat

Memory

VRAM Requirements

QualityQuantizationVRAMFile Size
fullQ8_013 GB11.5 GB
recommendedQ6_K10 GB8.8 GB
recommendedQ5_K_M8.5 GB7.6 GB
efficientQ4_K_M7.2 GB6.4 GB
compressedQ3_K_M5.8 GB5.2 GB
Scaling

Context Length Impact

KV cache VRAM at Q6_K quality. Longer context = more memory.

ContextKV CacheTotal VRAM
2K205 MB10.2 GB
4K410 MB10.4 GB
8K819 MB10.8 GB
16K1.5 GB11.5 GB
32K3.1 GB13.1 GB
64K6.1 GB16.1 GB
128K12.3 GB22.3 GB

Compatible GPUs

34 devices
NVIDIA Grace Blackwell Ultra GB300Q8_0260 tok/sExcellent
Apple M4 Max (128GB Unified)Q8_042 tok/sExcellent
Apple M4 Max (36GB Unified)Q8_042 tok/sExcellent
Apple M4 Max (64GB Unified)Q8_042 tok/sExcellent
Apple M4 Ultra (192GB)Q8_063 tok/sExcellent
NVIDIA GeForce RTX 4070 Ti SuperQ8_055 tok/sExcellent
NVIDIA GeForce RTX 4080 SuperQ8_068 tok/sExcellent
NVIDIA GeForce RTX 4090Q8_095 tok/sExcellent
NVIDIA GeForce RTX 5080Q8_0105 tok/sExcellent
NVIDIA GeForce RTX 5090Q8_0130 tok/sExcellent
NVIDIA RTX PRO 6000 BlackwellQ8_089 tok/sExcellent
NVIDIA RTX PRO 6000 Blackwell Max-QQ8_082 tok/sExcellent
Apple M4 (16GB Unified)Q8_014 tok/sGood
Apple M4 Pro (24GB Unified)Q8_028 tok/sGood
Apple M4 Pro (48GB)Q8_030 tok/sGood
NVIDIA GeForce RTX 4060 Ti 16GBQ6_K38 tok/sGood
NVIDIA GeForce RTX 4070 SuperQ6_K48 tok/sGood
NVIDIA RTX 4090 Laptop (150-175W)Q6_K32 tok/sGood
AMD Radeon RX 7900 XTXQ8_082 tok/sGood
AMD Radeon Pro W7900Q8_032 tok/sGood
AMD Radeon RX 9070Q6_K66 tok/sGood
Apple M1 Pro (16GB Unified)Q8_014 tok/sGood
Apple M2 Pro (16GB Unified)Q8_016 tok/sGood
NVIDIA GeForce RTX 5060 Ti 16GBQ6_K43 tok/sGood
NVIDIA GeForce RTX 3060 12GBQ4_K_M22 tok/sAcceptable
Apple M2 (16GB Unified)Q8_07 tok/sAcceptable
Apple M3 (16GB Unified)Q8_08 tok/sAcceptable
AMD Radeon RX 9060 XT 16GBQ6_K33 tok/sAcceptable
AMD Radeon RX 7600Q3_K_MMarginal
Apple M1 (16GB Unified)Q8_04 tok/sMarginal
Apple M1 (8GB Unified)Q8_0Not viable
Apple M2 (8GB Unified)Q8_0Not viable
Apple M3 (8GB Unified)Q8_0Not viable
AMD Radeon RX 9060 XT 8GBQ6_KNot viable

Showing 34 of 34 entries

Builder Context

Llama 3.2 11B Vision is commonly used with LM Studio, Ollama, Open WebUI.

FAQ

Frequently Asked Questions

How much VRAM does Llama 3.2 11B Vision need?
Llama 3.2 11B Vision requires 10 GB VRAM at recommended quality (Q6_K). At lower quality settings, it can fit in as little as 5.8 GB.
What is the best GPU for Llama 3.2 11B Vision?
The NVIDIA Grace Blackwell Ultra GB300 delivers the best performance for Llama 3.2 11B Vision, achieving 260 tok/s at Q8_0 with an excellent rating.
Can I run Llama 3.2 11B Vision on an RTX 4060 Ti?
Yes. On the NVIDIA GeForce RTX 4060 Ti 16GB, Llama 3.2 11B Vision runs at 38 tok/s (Q6_K, good).
What quantization should I use for Llama 3.2 11B Vision?
For the best quality, use Q6_K (10 GB VRAM). If your GPU has limited VRAM, Q3_K_M (5.8 GB) is the most efficient option with acceptable quality.
All models

Data confidence: estimated. Source

VRAM requirements are calculated from model parameters and may vary by inference engine, context length, and batch size. Performance estimates are based on community benchmarks and should be verified for your specific configuration.Llama is a trademark of its respective owner. OwnRig is not affiliated with or endorsed by the model creator.