Mixtral 8x7B Instruct
Mixtral · Apache 2.0
Mixture of Experts: 46.7B total parameters, 12.9B active per token.
Mixture-of-Experts model: 46.7B total params but only ~12.9B active per token. Quality closer to a 13B dense model with inference speed to match. Needs more VRAM for the full weight set though.
- Parameters
- 46.7B
- Architecture
- MoE (12.9B active)
- Context
- 32,768 tokens
- Released
- 2024-01-08
- Engines
- llama.cpp, ollama, vLLM
Parameters
46.7B
VRAM
31.4 GB
Context
32K
Formats
4
GPUs
19
Mixtral 8x7B Instruct (46.7B) requires 31.4 GB VRAM at recommended quality (Q5_K_M). On NVIDIA RTX PRO 6000 Blackwell, expect approximately 125 tok/s at Q5_K_M. For the best experience, High-End Home AI Server ($3,842) is recommended.
Source: OwnRig methodology
31.4 GB
Q5_K_M
28 GB
32K tokens
Chat
VRAM Requirements
| Quality | Quantization | VRAM | File Size |
|---|---|---|---|
| recommended | Q5_K_M | 31.4 GB | 28 GB |
| efficient | Q4_K_M | 26.2 GB | 23.4 GB |
| compressed | Q3_K_M | 21 GB | 18.2 GB |
| compressed | Q2_K | 16.4 GB | 14 GB |
Context Length Impact
KV cache VRAM at Q5_K_M quality. Longer context = more memory.
| Context | KV Cache | Total VRAM |
|---|---|---|
| 2K | 410 MB | 31.8 GBexceeds 24 GB |
| 4K | 819 MB | 32.2 GBexceeds 24 GB |
| 8K | 1.5 GB | 32.9 GBexceeds 24 GB |
| 16K | 3.1 GB | 34.5 GBexceeds 24 GB |
| 32K | 6.1 GB | 37.5 GBexceeds 24 GB |
Compatible GPUs
19 devices| NVIDIA Grace Blackwell Ultra GB300 | Q5_K_M | 100 tok/s | Excellent |
| NVIDIA RTX PRO 6000 Blackwell | Q5_K_M | 125 tok/s | Excellent |
| NVIDIA RTX PRO 6000 Blackwell Max-Q | Q5_K_M | 115 tok/s | Excellent |
| Apple M4 Max (36GB Unified) | Q4_K_M | 20 tok/s | Good |
| Apple M4 Max (64GB Unified) | Q5_K_M | 18 tok/s | Good |
| NVIDIA GeForce RTX 4090 | Q3_K_M | 35 tok/s | Good |
| AMD Radeon RX 7900 XTX | Q3_K_M | 30 tok/s | Good |
| AMD Radeon Pro W7900 | Q5_K_M | 19 tok/s | Good |
| AMD Radeon RX 9070 | Q2_K | 4 tok/s | Marginal |
| AMD Radeon RX 9060 XT 16GB | Q2_K | 2 tok/s | Marginal |
| Apple M3 Pro (18GB Unified) | Q2_K | – | Not viable |
| NVIDIA GeForce RTX 3080 10GB | Q2_K | – | Not viable |
| NVIDIA GeForce RTX 4060 8GB | Q4_K_M | – | Not viable |
| NVIDIA RTX 4060 Laptop (40-60W) | Q4_K_M | – | Not viable |
| NVIDIA RTX 4070 Laptop (80-115W) | Q4_K_M | – | Not viable |
| NVIDIA GeForce RTX 4070 Ti 12GB | Q4_K_M | – | Not viable |
| AMD Radeon RX 7600 | Q2_K | – | Not viable |
| AMD Radeon RX 9060 XT 8GB | Q2_K | – | Not viable |
| NVIDIA GeForce RTX 5060 8GB | Q4_K_M | – | Not viable |
Showing 19 of 19 entries
Recommended Builds
Complete PC builds that can run Mixtral 8x7B Instruct.
Extreme AI Workstation
Dual GPUs that run the biggest AI models at a smart price
Runs 8 models
High-End AI Workstation
Chat, generate images, and code with AI, all at once
Runs 8 models
High-End Home AI Server
Your household's private AI: chatbots, code tools, and more
Runs 12 models
Frequently Asked Questions
- How much VRAM does Mixtral 8x7B Instruct need?
- Mixtral 8x7B Instruct requires 31.4 GB VRAM at recommended quality (Q5_K_M). At lower quality settings, it can fit in as little as 16.4 GB.
- What is the best GPU for Mixtral 8x7B Instruct?
- The NVIDIA RTX PRO 6000 Blackwell delivers the best performance for Mixtral 8x7B Instruct, achieving 125 tok/s at Q5_K_M with an excellent rating.
- What quantization should I use for Mixtral 8x7B Instruct?
- For the best quality, use Q5_K_M (31.4 GB VRAM). If your GPU has limited VRAM, Q2_K (16.4 GB) is the most efficient option with acceptable quality.
- Is Mixtral 8x7B Instruct good for coding?
- Mixtral 8x7B Instruct supports coding use cases. For the best coding experience, pair it with an embedding model for local RAG.
Data confidence: verified. 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.Mixtral is a trademark of its respective owner. OwnRig is not affiliated with or endorsed by the model creator.