How much VRAM does Devstral 2 123B Instruct need?

Devstral 2 123B Instruct (123B parameters) requires approximately 98.2 GB of memory with Q4_K_M quantization.

What is the best quantization for Devstral 2 123B Instruct?

The recommended quantization for Devstral 2 123B Instruct is Q4_K_M, which balances quality and memory efficiency.

Can it run?

No, Devstral 2 123B Instruct requires more memory than RX 6950 XT 16GB provides.

FWon't run

Too heavy

Using Q4_K_M in vLLM

Capabilities:

Fit status

Too heavy

Decode

4.5 tok/s

TTFT

43480 ms

Safe context

Memory

98.2 GB / 16.0 GB

Weights75.0 GB

KV Cache19.2 GB

Runtime2.4 GB

Headroom1.6 GB

Workload	Grade	Fit	Decode	TTFT	Context
Agentic Coding	F	Too heavy	4.5 tok/s	63244 ms	4K
Chat	F	Too heavy	4.5 tok/s	23716 ms	4K
Coding	F	Too heavy	4.5 tok/s	43480 ms	4K
RAG	F	Too heavy	4.5 tok/s	79055 ms	4K
Reasoning	F	Too heavy	5.5 tok/s	41394 ms	4K

How Devstral 2 123B Instruct (123B params) fits at each quantization level on RX 6950 XT 16GB (16.0 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	48.0 GB	Low	F0
Q3_K_S	3	60.3 GB	Low	F0
NVFP4	4

Upgrade options

MacBook Pro M3 Max 128GBBest value

D3 tok/s decode

~$2,499 MSRP

AMD Instinct MI350X 288GBBudget pick

C77.8 tok/s decode

~$8,000 MSRP