Llama 3 12B — Hardware Requirements & GPU Compatibility
Chat
Llama 3 12B is a 11.5B-parameter open language model from mlabonne in the Llama 3 family. It supports a context window of up to 8,192 tokens. At Q4_K_M it needs about 7.61 GB of VRAM — see which GPUs and Macs can run it below.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 2K)Best on 24 GB (Q4_K_M · ~86 tok/s · 8K)Macs from 16 GB (Q4_K_M · ~18 tok/s)
Other
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Llama 3 12B
One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.
Recommended
vllm serve mlabonne/Llama-3-12B --max-model-len 8192vLLM repo with its own quantisation
Download: Weights on Hugging Face
Ways to runOther ways
each opens step-by-stepRun in cloud
Runs on RTX 3080 Ti (12 GB) (3 GB headroom) · Q4_K_M
- Generation speed
- ~78 tok/s
- generation speed
- Cost per 1M output tokens
- $0.44
- per 1M output tokens
How Much VRAM Does Llama 3 12B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 5.6 GB | — | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 8K context · ~117 tok/s | |
| 6.3 GB | — | 3-bit medium — 3-bit medium quantization | Fits: 8K context · ~104 tok/s | |
| 7.6 GB | — | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 8K context · ~86 tok/s | |
| 8.9 GB | — | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 8K context · ~74 tok/s | |
| 10.2 GB | — | 6-bit, very good — 6-bit quantization, very good quality | Fits: 8K context · ~64 tok/s | |
| 12.2 GB | — | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 8K context · ~54 tok/s | |
| 23.7 GB | — | 16-bit · bfloat — Brain floating point 16 — preferred for training | Fits: 2K context · ~28 tok/s |
Selected Q4_K_M
4-bit medium quantization8.8GBat 8K context
- Weights
- 6.9 GB
- KV cache
- 1.6 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 15.2 GB free · ~86 tok/s
≥50 tok/s needs ≥585 GB/s · RTX 5090 ~153, RTX 3090 Ti ~86
Which Hardware Can Run Llama 3 12B?
Q4_K_M · 7.6 GBLlama 3 12B (Q4_K_M) requires 7.6 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 10+ GB is recommended. Using the full 8K context window can add up to 1.2 GB, bringing total usage to 8.8 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 3080. 151 devices with unified memory can run Llama 3 12B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), NVIDIA Jetson Orin Nano 8GB (Super).
12 GB
12 GB tier →fits · 3.2 GB free · up to 8K ctx
10 GPUs
RTX 3080 Ti~78Simulate output on RTX 3080 TiRTX 5070~57Simulate output on RTX 5070RTX 4070~43Simulate output on RTX 407020 GB
fits · 11.2 GB free · up to 8K ctx
1 GPU
RX 7900 XT~63Simulate output on RX 7900 XT32–40 GB
fits · 23.2 GB free · up to 8K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~153Simulate output on RTX 5090M5 Max36 GB~42Simulate output on Mac Studio M5 Max (36 GB)M532 GB~14Simulate output on MacBook Air 13" M5 (32 GB)Orin~18Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~12Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Alternatives to Llama 3 12B
Llama 3 family · 65 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Gemma 3 12B ITGoogle | 74.2 | 10.9 GB | ~82 | same size |
| Gemma 4 12BGoogle | — | 9 GB | ~80 | same size |
| Qwen2.5 Omni 7BAlibaba | — | 9.9 GB | ~93 | same size |
| Mistral Nemo Base 2407Mistral AI | — | 11 GB | ~81 | same size |
| Falcon 11BTII UAE | — | 8.5 GB | ~89 | same size |
| Bielik 11B V3.0 Instructspeakleash | — | 27.4 GB | —needs > 24 GB | same size |
Specifications
- Publisher
- mlabonne
- Family
- Llama 3
- Parameters
- 11.5B
- Architecture
- LlamaForCausalLM
- Context Length
- 8,192 tokens
- Vocabulary Size
- 128,256
- Release Date
- 2024-04-18
- License
- Other
Frequently Asked Questions
How much VRAM does Llama 3 12B need?
Llama 3 12B requires 7.6 GB of VRAM at Q4_K_M, or 23.7 GB at BF16. Full 8K context adds up to 1.2 GB (8.8 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 11.5B × 4.8 bits ÷ 8 = 6.9 GB
KV Cache + Overhead ≈ 0.7 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 1.9 GB (at full 8K context)
VRAM usage by quantization
What's the best quantization for Llama 3 12B?
For Llama 3 12B, Q4_K_M (7.6 GB) offers the best balance of quality and VRAM usage. Q5_K_M (8.9 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 5.6 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Llama 3 12B on a Mac?
Yes — Mac Mini M2 Pro (16 GB) and 133 other Macs can run Llama 3 12B. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 7.6 GB of usable unified memory (RAM minus macOS overhead).
Can I run Llama 3 12B locally?
Yes — Llama 3 12B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Llama 3 12B?
At Q4_K_M, Llama 3 12B can reach ~631 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~86 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 7.61 × 0.65 = ~683 tok/s
Estimated speed at Q4_K_M (7.6 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Llama 3 12B?
At Q4_K_M, the download is about 6.91 GB. The full-precision BF16 version is 23.04 GB. The smallest option (Q2_K) is 4.90 GB.
Which GPUs can run Llama 3 12B?
52 consumer GPUs can run Llama 3 12B at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT, AMD Radeon RX 7600. 36 GPUs have plenty of headroom for comfortable inference.
Which devices can run Llama 3 12B?
151 devices with unified memory can run Llama 3 12B at Q4_K_M (7.6 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.