How much VRAM does Mixtral 8x22B need?

Mixtral 8x22B (141B parameters) requires approximately 106.8 GB of memory with Q4_K_M quantization.

What is the best quantization for Mixtral 8x22B?

The recommended quantization for Mixtral 8x22B is Q4_K_M, which balances quality and memory efficiency.

Can it run?

Yes, MacBook Pro M3 Max 128GB can run Mixtral 8x22B with a D grade (Very compromised). Expected decode speed: 5.6 tok/s.

DPoor

Very compromised

Using Q4_K_M in llama.cpp

Capabilities:

Fit status

Very compromised

Decode

5.6 tok/s

TTFT

34819 ms

Safe context

14K

Memory

106.8 GB / 92.2 GB

Offload

10%

Weights86.0 GB

KV Cache6.1 GB

Runtime0.9 GB

Headroom13.8 GB

Workload	Grade	Fit	Decode	TTFT	Context
Agentic Coding	F	Too heavy	5.4 tok/s	52606 ms	26K
Chat	D	Very compromised (needs ~9.6 GB host RAM)	5.2 tok/s	20413 ms	7K
Coding	D	Very compromised	5.6 tok/s	34819 ms	14K
RAG	F	Too heavy	5.4 tok/s	65758 ms	26K
Reasoning	D	Very compromised (needs ~11.8 GB host RAM)	5.1 tok/s	44579 ms

How Mixtral 8x22B (141B params) fits at each quantization level on MacBook Pro M3 Max 128GB (92.2 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	55.0 GB	Low	C42
Q3_K_SBest for your GPU	3	69.1 GB	Low	C45

Ollama

ollama run mixtral-8x22b

HuggingFace

huggingface-cli download mixtral-8x22b

Upgrade options

AMD Instinct MI350X 288GBBudget pick

C130.3 tok/s decode

~$8,000 MSRP

AMD Instinct MI300X 192GBBest value

C92.2 tok/s decode

~$15,000 MSRP

NVIDIA H200 141GBBiggest leap