# Llama 3.3 70B Instruct: hardware requirements
> Llama 3.3 70B Instruct (Meta) is a 70.6B dense model with a 128K-token context window, released under the Llama 3.3 Community License. It needs about 47GB of memory at Q4_K_M with 8K context.
Llama 3.3 70B was Meta's last dense flagship and, for most of 2025, the model that justified buying a second 24GB GPU. It is still a capable writer and generalist, but 2026 releases such as Qwen3.8 27B and Qwen3.5 122B-A10B beat it on reasoning and coding while needing less memory or running much faster. At Q4_K_M it needs about 47GB with 8K context, so plan on 48GB of VRAM (two 24GB cards with a short context, or one 48GB workstation card) or a Mac with 64GB or more of unified memory. Generation is slow on Macs because every token reads all 70B parameters.

## Recommended hardware
- **Minimum GPU:** RTX A6000 (48GB), Q3_K_M, ~16 tokens/sec
- **Best value:** RTX 6000 Ada (48GB), Q3_K_M, ~22 tokens/sec
- **Fastest single GPU:** RTX 6000 Ada (48GB), Q3_K_M, ~22 tokens/sec
- **Multi-GPU:** 3× RTX 3090 (72GB), Q4_K_M, ~15 tokens/sec
- **Multi-GPU:** 3× RTX 4090 (72GB), Q4_K_M, ~18 tokens/sec
- **Multi-GPU:** 2× RTX 5090 (64GB), Q4_K_M, ~31 tokens/sec
- **Mac / unified memory:** Ryzen AI Max+ 395 (128GB), Q8_0, ~2.6 tokens/sec

## Memory by quantization
| Quant | Weights | Total @ 8K ctx | Total @ 32K ctx |
| --- | --- | --- | --- |
| Q2_K | 26.3GB | 30.4GB | 38.4GB |
| Q3_K_M | 34.2GB | 38.5GB | 46.6GB |
| Q4_K_M | 42.5GB | 47.1GB | 55.1GB |
| Q5_K_M | 50.0GB | 54.7GB | 62.8GB |
| Q6_K | 57.9GB | 62.9GB | 71.0GB |
| Q8_0 | 75.0GB | 80.5GB | 88.6GB |
| BF16 | 141.1GB | 148.7GB | 156.7GB |

## GPUs that run it fully in memory (fastest first)
| GPU | Memory | Quant | Speed | Price |
| --- | --- | --- | --- | --- |
| RTX 6000 Ada | 48GB | Q3_K_M | ~22 tok/s | ~$4,750 used |
| RTX PRO 6000 | 96GB | Q8_0 | ~19 tok/s | ~$16,000 |
| RTX PRO 5000 72GB | 72GB | Q6_K | ~18 tok/s | ~$9,100 |
| RTX A6000 | 48GB | Q3_K_M | ~16 tok/s | ~$4,400 used |
| M5 Ultra 512GB | 512GB | Q8_0 | ~8.9 tok/s | n/a |
| M5 Ultra 256GB | 256GB | Q8_0 | ~8.9 tok/s | ~$9,499 |
| M3 Ultra 512GB | 512GB | Q8_0 | ~6.0 tok/s | $9,499 MSRP |
| M5 Max 128GB | 128GB | Q8_0 | ~5.3 tok/s | ~$5,100 |
| M4 Max 128GB | 128GB | Q8_0 | ~4.7 tok/s | $3,499 MSRP |
| DGX Spark | 128GB | Q8_0 | ~2.7 tok/s | ~$6,950 |
| Ryzen AI Max+ PRO 495 | 192GB | Q8_0 | ~2.7 tok/s | ~$6,800 |
| Ryzen AI Max+ 395 | 128GB | Q8_0 | ~2.6 tok/s | ~$3,300 |

## How to run it
- Ollama: `ollama run llama3.3:70b`
- llama.cpp: `llama-server -hf bartowski/Llama-3.3-70B-Instruct-GGUF:Q4_K_M -c 8192 -ngl 99`
- MLX (Mac): `mlx_lm.chat --model mlx-community/Llama-3.3-70B-Instruct-4bit`
- Weights: https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct

## Strengths
- Strong general writing and instruction following for a non-reasoning model
- Dense weights quantize predictably; Q4_K_M holds up well
- Excellent runtime support and a large ecosystem of fine-tunes

## Weaknesses
- Needs about 47GB at Q4_K_M, more than any single consumer GPU
- Slow on bandwidth-limited hardware because all 70B parameters are read for every token
- Outclassed on reasoning and coding by 2026 models a third of its size
- Text only, with no reasoning mode
Full page: https://llamabuilds.ai/models/llama-3-3-70b · Hardware guide: https://llamabuilds.ai/models/llama-3-3-70b/requirements

---
Source: https://llamabuilds.ai/models/llama-3-3-70b · Updated October 14, 2026 · Llama Builds. Speeds marked ~ are estimates from our bandwidth model (https://llamabuilds.ai/methodology); measured results are labeled as such.