How much VRAM does Mixtral 8x22B need?

Mixtral 8x22B (141B parameters) requires approximately 97.8 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?

No, Mixtral 8x22B requires more memory than RTX 6000 Ada 48GB provides.

FWon't run

Too heavy

Using Q4_K_M in llama.cpp

Capabilities:

Fit status

Too heavy

Decode

15.4 tok/s

TTFT

12558 ms

Safe context

Memory

97.8 GB / 48.0 GB

Weights86.0 GB

KV Cache6.1 GB

Runtime0.9 GB

Headroom4.8 GB

Workload	Grade	Fit	Decode	TTFT	Context
Agentic Coding	F	Too heavy	17.6 tok/s	16039 ms	15K
Chat	F	Too heavy	17.6 tok/s	6014 ms	4K
Coding	F	Too heavy	15.4 tok/s	12558 ms	8K
RAG	F	Too heavy	17.6 tok/s	20048 ms	15K
Reasoning	F	Too heavy	17.6 tok/s	13031 ms	8K

How Mixtral 8x22B (141B params) fits at each quantization level on RTX 6000 Ada 48GB (48.0 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	55.0 GB	Low	F0
Q3_K_S	3	69.1 GB	Low	F0
NVFP4	4

Upgrade options

MacBook Pro M3 Max 128GBBest value

D5.1 tok/s decode

~$2,499 MSRP

AMD Instinct MI350X 288GBBudget pick

C130.3 tok/s decode

~$8,000 MSRP

NVIDIA H200 141GBBiggest leap