How much VRAM does Mixtral 8x22B need?

Mixtral 8x22B (141B parameters) requires approximately 120.7 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, Mac Studio M3 Ultra 256GB can run Mixtral 8x22B with a C grade (Runs well). Expected decode speed: 11.9 tok/s.

CUsable

Runs well

Using Q4_K_M in llama.cpp

Capabilities:

Fit status

Runs well

Decode

11.9 tok/s

TTFT

16305 ms

Safe context

24K

Memory

120.7 GB / 184.3 GB

Weights86.0 GB

KV Cache6.1 GB

Runtime0.9 GB

Headroom27.6 GB

Workload	Grade	Fit	Decode	TTFT	Context
Agentic Coding	C	Runs well	11.9 tok/s	23717 ms	47K
Chat	C	Runs well	11.9 tok/s	8894 ms	13K
Coding	C	Runs well	11.9 tok/s	16305 ms	24K
RAG	C	Runs well	11.9 tok/s	29646 ms	47K
Reasoning	C	Runs well	11.9 tok/s	19270 ms	24K

How Mixtral 8x22B (141B params) fits at each quantization level on Mac Studio M3 Ultra 256GB (184.3 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	55.0 GB	Low	D36
Q3_K_S	3	69.1 GB	Low	D38
NVFP4	4

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

H100 NVL 188GBBiggest leap