How much VRAM does Mixtral 8x22B need?

Mixtral 8x22B (141B parameters) requires approximately 101.0 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 NVIDIA H100 PCIe 80GB provides.

FWon't run

Too heavy

Using Q4_K_M in llama.cpp

Capabilities:

Fit status

Too heavy

Decode

32.1 tok/s

TTFT

6028 ms

Safe context

13K

Memory

101.0 GB / 80.0 GB

Weights86.0 GB

KV Cache6.1 GB

Runtime0.9 GB

Headroom8.0 GB

Workload	Grade	Fit	Decode	TTFT	Context
Agentic Coding	F	Too heavy	37.5 tok/s	7515 ms	24K
Chat	F	Too heavy	37.5 tok/s	2818 ms	7K
Coding	F	Too heavy	32.1 tok/s	6028 ms	13K
RAG	F	Too heavy	37.5 tok/s	9394 ms	24K
Reasoning	F	Too heavy	37.5 tok/s	6106 ms	13K

How Mixtral 8x22B (141B params) fits at each quantization level on NVIDIA H100 PCIe 80GB (80.0 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_KBest for your GPU	2	55.0 GB	Low	C44
Q3_K_S	3	69.1 GB	Low	C45

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