Devstral-Small-2-24B-Instruct-2512

Mistral · 24.0B · ctx 256K · apache-2.0 · HuggingFace

Devstral-Small-2-24B-Instruct-2512 (Mistral) is a 24.0B parameters open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 19.0 GB VRAM for Q4 (weights 13.2 GB + KV 4.8 GB + 1 GB overhead) and 53.8 GB for FP16. About 22 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 212.1 tok/s (estimated); MI300X 192GB ≈ 140.5 tok/s (estimated); H200 141GB SXM5 ≈ 127.3 tok/s (estimated); H100 80GB SXM5 ≈ 88.8 tok/s (estimated); Gaudi 3 128GB ≈ 86.9 tok/s (estimated). Use the table below for VRAM fit and tok/s per dollar (list MSRP). Estimates follow memory-bandwidth math; measured rows override when present. See methodology for details. Methodology.

85 hardware configs — sorted by tok/s. Default fit context 8K.

Hardware	VRAM	Quant	tok/s	Fits?	tok/s/$
B200 192GB SXM	192	Q4	212.1 est.	Yes	0.005
MI300X 192GB	192	Q4	140.5 est.	Yes	0.009
H200 141GB SXM5	141	Q4	127.3 est.	Yes	0.004
H100 80GB SXM5	80	Q4	88.8 est.	Yes	0.003
Gaudi 3 128GB	128	Q4	86.9 est.	Yes	0.007
B200 192GB SXM	192	FP16	58.3 est.	Yes	0.001
A100 80GB SXM4	80	Q4	54.1 est.	Yes	0.004
RTX 5080 16GB	16	Q4	51.6 est.	No	0.052
RTX 5070 Ti 16GB	16	Q4	48.1 est.	No	0.064
RTX 5090 32GB	32	Q4	47.5 est.	Yes	0.024
RTX 4080 Super 16GB	16	Q4	39.5 est.	No	0.04
MI300X 192GB	192	FP16	38.6 est.	Yes	0.003
RTX 4070 Ti Super 16GB	16	Q4	36.1 est.	No	0.045
H200 141GB SXM5	141	FP16	35.0 est.	Yes	0.001
RX 9070 16GB	16	Q4	34.4 est.	No	0.063
RX 9070 XT 16GB	16	Q4	34.4 est.	No	0.057
RX 7800 XT 16GB	16	Q4	33.5 est.	No	0.067
RX 7900 GRE 16GB	16	Q4	30.9 est.	No	0.056
Arc A770 16GB	16	Q4	30.1 est.	No	0.086
RTX 4090 24GB	24	Q4	26.7 est.	Yes	0.017
RTX 4090D 24GB (China)	24	Q4	26.7 est.	Yes	0.017
RTX 3090 Ti 24GB	24	Q4	26.7 est.	Yes	0.013
RX 7900 XTX 24GB	24	Q4	25.5 est.	Yes	0.026
RTX 6000 Ada 48GB	48	Q4	25.5 est.	Yes	0.004
RTX 3090 24GB	24	Q4	24.8 est.	Yes	0.017
H100 80GB SXM5	80	FP16	24.4 est.	Yes	0.001
RTX 5060 Ti 16GB	16	Q4	24.1 est.	No	0.056
Gaudi 3 128GB	128	FP16	23.9 est.	Yes	0.002
L40S 48GB	48	Q4	22.9 est.	Yes	0.003
Mac Studio M3 Ultra 96GB	96	Q4	21.7 est.	Yes	0.005
Mac Studio M2 Ultra 192GB	192	Q4	21.2 est.	Yes	0.003
RX 7900 XT 20GB	20	Q4	21.2 est.	Yes	0.024
RTX A6000 48GB	48	Q4	20.4 est.	Yes	0.005
RTX 5070 12GB	12	Q4	20.3 est.	No	0.037
T4 16GB	16	Q4	17.2 est.	No	0.009
RTX 4070 Ti 12GB	12	Q4	16.7 est.	No	0.021
RTX 3080 10GB	10	Q4	15.9 est.	No	0.023
RTX 6000 Ada 48GB	48	FP16	15.9 est.	No	0.002
RTX 4060 Ti 16GB	16	Q4	15.5 est.	No	0.031
RX 7600 XT 16GB	16	Q4	15.5 est.	No	0.047
RTX 4070 12GB	12	Q4	15.2 est.	No	0.025
RTX 4070 Super 12GB	12	Q4	15.2 est.	No	0.025
A100 80GB SXM4	80	FP16	14.9 est.	Yes	0.001
MacBook Pro M4 Max 128GB	128	Q4	14.5 est.	Yes	0.003
L40S 48GB	48	FP16	14.3 est.	No	0.002
Arc B580 12GB	12	Q4	13.8 est.	No	0.055
RTX 5090 32GB	32	FP16	13.2 est.	No	0.007
RX 7700 XT 12GB	12	Q4	13.1 est.	No	0.029
RTX A6000 48GB	48	FP16	12.7 est.	No	0.003
RTX 3060 12GB	12	Q4	10.9 est.	No	0.033
MacBook Pro M3 Max 128GB	128	Q4	10.6 est.	Yes	0.003
Arc B570 10GB	10	Q4	8.0 est.	No	0.037
Mac mini M4 Pro 64GB	64	Q4	7.2 est.	Yes	0.003
RTX 5060 8GB	8	Q4	6.0 est.	No	0.02
RTX 5060 Ti 8GB	8	Q4	6.0 est.	No	0.016
RTX 3070 8GB	8	Q4	6.0 est.	No	0.012
Mac Studio M3 Ultra 96GB	96	FP16	6.0 est.	Yes	0.002
Mac Studio M2 Ultra 192GB	192	FP16	5.8 est.	Yes	0.001
RTX 4090 24GB	24	FP16	4.2 est.	No	0.003
RTX 4090D 24GB (China)	24	FP16	4.2 est.	No	0.003
RTX 3090 Ti 24GB	24	FP16	4.2 est.	No	0.002
RX 7900 XTX 24GB	24	FP16	4.0 est.	No	0.004
MacBook Pro M4 Max 128GB	128	FP16	4.0 est.	Yes	0.001
RTX 3090 24GB	24	FP16	3.9 est.	No	0.003
RTX 4060 Ti 8GB	8	Q4	3.9 est.	No	0.01
RX 7600 8GB	8	Q4	3.9 est.	No	0.014
RTX 4060 8GB	8	Q4	3.7 est.	No	0.012
Mac mini M4 32GB	32	Q4	3.2 est.	Yes	0.003
MacBook Pro M3 Max 128GB	128	FP16	2.9 est.	Yes	0.001
RX 7900 XT 20GB	20	FP16	2.3 est.	No	0.003
Mac mini M4 Pro 64GB	64	FP16	2.0 est.	Yes	0.001
RTX 5080 16GB	16	FP16	1.8 est.	No	0.002
RTX 5070 Ti 16GB	16	FP16	1.7 est.	No	0.002
RTX 4080 Super 16GB	16	FP16	1.4 est.	No	0.001
RTX 4070 Ti Super 16GB	16	FP16	1.2 est.	No	0.002
RX 9070 16GB	16	FP16	1.2 est.	No	0.002
RX 9070 XT 16GB	16	FP16	1.2 est.	No	0.002
RX 7800 XT 16GB	16	FP16	1.1 est.	No	0.002
RX 7900 GRE 16GB	16	FP16	1.1 est.	No	0.002
Arc A770 16GB	16	FP16	1.0 est.	No	0.003
Mac mini M4 32GB	32	FP16	0.9 est.	No	0.001
RTX 5060 Ti 16GB	16	FP16	0.8 est.	No	0.002
T4 16GB	16	FP16	0.6 est.	No	0.0
RTX 4060 Ti 16GB	16	FP16	0.5 est.	No	0.001
RX 7600 XT 16GB	16	FP16	0.5 est.	No	0.002

Fit checks

• RTX 5090 32GB — fits · ≈47.5 tok/s Q4 est.
• RTX 3090 Ti 24GB — fits · ≈26.7 tok/s Q4 est.
• RTX 4090 24GB — fits · ≈26.7 tok/s Q4 est.
• RTX 5080 16GB — does not fit · ≈51.6 tok/s Q4 est.
• RTX 6000 Ada 48GB — fits · ≈25.5 tok/s Q4 est.
• RX 7900 XTX 24GB — fits · ≈25.5 tok/s Q4 est.
• RTX 3090 24GB — fits · ≈24.8 tok/s Q4 est.
• RTX 5070 Ti 16GB — does not fit · ≈48.1 tok/s Q4 est.
• L40S 48GB — fits · ≈22.9 tok/s Q4 est.
• Mac Studio M3 Ultra 96GB — fits · ≈21.7 tok/s Q4 est.
• Mac Studio M2 Ultra 192GB — fits · ≈21.2 tok/s Q4 est.
• RX 7900 XT 20GB — fits · ≈21.2 tok/s Q4 est.

12 popular retail GPUs; every row in the table above also links its Fits? verdict to the full fit check.