Mixtral-8x7B-Instruct-v0.1 — which GPUs fit

Mistral · 46.7B (active 12.9B) · MoE · ctx 32K · apache-2.0 · HuggingFace

Mixtral-8x7B-Instruct-v0.1 (Mistral) is a 46.7B parameters MoE (12.9B active per token) open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 29.27 GB VRAM for Q4 (weights 25.69 GB + KV 2.58 GB + 1 GB overhead) and 96.98 GB for FP16. About 16 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 394.6 tok/s (estimated); MI300X 192GB ≈ 261.5 tok/s (estimated); H200 141GB SXM5 ≈ 236.8 tok/s (estimated); H100 80GB SXM5 ≈ 165.3 tok/s (estimated); Gaudi 3 128GB ≈ 161.6 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.

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

Hardware	VRAM	Quant	tok/s	Fits?	tok/s/$
B200 192GB SXM	192	Q4	394.6 est.	Yes	0.01
MI300X 192GB	192	Q4	261.5 est.	Yes	0.017
H200 141GB SXM5	141	Q4	236.8 est.	Yes	0.007
H100 80GB SXM5	80	Q4	165.3 est.	Yes	0.006
Gaudi 3 128GB	128	Q4	161.6 est.	Yes	0.013
B200 192GB SXM	192	FP16	108.5 est.	Yes	0.003
A100 80GB SXM4	80	Q4	100.6 est.	Yes	0.007
RTX 4090 24GB	24	Q4	95.5 est.	No	0.06
RTX 4090D 24GB (China)	24	Q4	95.5 est.	No	0.06
RTX 3090 Ti 24GB	24	Q4	95.5 est.	No	0.048
RX 7900 XTX 24GB	24	Q4	91.0 est.	No	0.091
RTX 3090 24GB	24	Q4	88.7 est.	No	0.059
H100 80GB SXM5	80	FP16	88.4 est.	No	0.003
RTX 5090 32GB	32	Q4	88.4 est.	Yes	0.044
MI300X 192GB	192	FP16	71.9 est.	Yes	0.005
H200 141GB SXM5	141	FP16	65.1 est.	Yes	0.002
A100 80GB SXM4	80	FP16	53.8 est.	No	0.004
RX 7900 XT 20GB	20	Q4	52.6 est.	No	0.059
RTX 6000 Ada 48GB	48	Q4	47.4 est.	Yes	0.007
Gaudi 3 128GB	128	FP16	44.4 est.	Yes	0.004
L40S 48GB	48	Q4	42.6 est.	Yes	0.005
RTX 5080 16GB	16	Q4	40.4 est.	No	0.04
Mac Studio M3 Ultra 96GB	96	Q4	40.4 est.	Yes	0.01
Mac Studio M2 Ultra 192GB	192	Q4	39.5 est.	Yes	0.006
RTX A6000 48GB	48	Q4	37.9 est.	Yes	0.008
RTX 5070 Ti 16GB	16	Q4	37.7 est.	No	0.05
Mac Studio M3 Ultra 96GB	96	FP16	31.1 est.	No	0.008
RTX 4080 Super 16GB	16	Q4	31.0 est.	No	0.031
RTX 4070 Ti Super 16GB	16	Q4	28.3 est.	No	0.035
RX 9070 16GB	16	Q4	27.0 est.	No	0.049
RX 9070 XT 16GB	16	Q4	27.0 est.	No	0.045
MacBook Pro M4 Max 128GB	128	Q4	26.9 est.	Yes	0.006
RX 7800 XT 16GB	16	Q4	26.3 est.	No	0.053
RX 7900 GRE 16GB	16	Q4	24.3 est.	No	0.044
Arc A770 16GB	16	Q4	23.6 est.	No	0.068
MacBook Pro M3 Max 128GB	128	Q4	19.7 est.	Yes	0.005
RTX 5060 Ti 16GB	16	Q4	18.9 est.	No	0.044
RTX 5070 12GB	12	Q4	15.9 est.	No	0.029
T4 16GB	16	Q4	13.5 est.	No	0.007
Mac mini M4 Pro 64GB	64	Q4	13.5 est.	Yes	0.006
RTX 4070 Ti 12GB	12	Q4	13.1 est.	No	0.016
RTX 3080 10GB	10	Q4	12.5 est.	No	0.018
RTX 4060 Ti 16GB	16	Q4	12.1 est.	No	0.024
RX 7600 XT 16GB	16	Q4	12.1 est.	No	0.037
RTX 4070 12GB	12	Q4	11.9 est.	No	0.02
RTX 4070 Super 12GB	12	Q4	11.9 est.	No	0.02
Mac Studio M2 Ultra 192GB	192	FP16	10.9 est.	Yes	0.002
Arc B580 12GB	12	Q4	10.8 est.	No	0.043
RX 7700 XT 12GB	12	Q4	10.2 est.	No	0.023
RTX 6000 Ada 48GB	48	FP16	9.1 est.	No	0.001
RTX 3060 12GB	12	Q4	8.5 est.	No	0.026
L40S 48GB	48	FP16	8.2 est.	No	0.001
RTX 5090 32GB	32	FP16	7.6 est.	No	0.004
MacBook Pro M4 Max 128GB	128	FP16	7.4 est.	Yes	0.002
RTX A6000 48GB	48	FP16	7.3 est.	No	0.002
Arc B570 10GB	10	Q4	6.3 est.	No	0.029
Mac mini M4 32GB	32	Q4	5.9 est.	Yes	0.006
MacBook Pro M3 Max 128GB	128	FP16	5.4 est.	Yes	0.001
RTX 5060 8GB	8	Q4	4.7 est.	No	0.016
RTX 5060 Ti 8GB	8	Q4	4.7 est.	No	0.012
RTX 3070 8GB	8	Q4	4.7 est.	No	0.009
Mac mini M4 Pro 64GB	64	FP16	4.6 est.	No	0.002
RTX 4060 Ti 8GB	8	Q4	3.0 est.	No	0.008
RX 7600 8GB	8	Q4	3.0 est.	No	0.011
RTX 4060 8GB	8	Q4	2.9 est.	No	0.01
Mac mini M4 32GB	32	FP16	0.5 est.	No	0.001