Qwen3.5-9B — which GPUs fit

Alibaba · 8.95B · ctx 256K · apache-2.0 · HuggingFace

Qwen3.5-9B (Alibaba) is a 8.95B parameters open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 7.71 GB VRAM for Q4 (weights 4.92 GB + KV 1.79 GB + 1 GB overhead) and 20.69 GB for FP16. About 44 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 568.8 tok/s (estimated); MI300X 192GB ≈ 376.8 tok/s (estimated); H200 141GB SXM5 ≈ 341.3 tok/s (estimated); H100 80GB SXM5 ≈ 238.2 tok/s (estimated); Gaudi 3 128GB ≈ 232.9 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.

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

Hardware	VRAM	Quant	tok/s	Fits?	tok/s/$
B200 192GB SXM	192	Q4	568.8 est.	Yes	0.014
MI300X 192GB	192	Q4	376.8 est.	Yes	0.025
H200 141GB SXM5	141	Q4	341.3 est.	Yes	0.01
H100 80GB SXM5	80	Q4	238.2 est.	Yes	0.008
Gaudi 3 128GB	128	Q4	232.9 est.	Yes	0.019
B200 192GB SXM	192	FP16	156.4 est.	Yes	0.004
A100 80GB SXM4	80	Q4	145.0 est.	Yes	0.01
RTX 5090 32GB	32	Q4	127.4 est.	Yes	0.064
MI300X 192GB	192	FP16	103.6 est.	Yes	0.007
RTX 5060 8GB	8	Q4	98.0 est.	No	0.328
RTX 5060 Ti 8GB	8	Q4	98.0 est.	No	0.259
RTX 3070 8GB	8	Q4	98.0 est.	No	0.196
H200 141GB SXM5	141	FP16	93.9 est.	Yes	0.003
RTX 4090 24GB	24	Q4	71.7 est.	Yes	0.045
RTX 4090D 24GB (China)	24	Q4	71.7 est.	Yes	0.045
RTX 3090 Ti 24GB	24	Q4	71.7 est.	Yes	0.036
RTX 5080 16GB	16	Q4	68.3 est.	Yes	0.068
RX 7900 XTX 24GB	24	Q4	68.3 est.	Yes	0.068
RTX 6000 Ada 48GB	48	Q4	68.3 est.	Yes	0.01
RTX 3090 24GB	24	Q4	66.6 est.	Yes	0.044
H100 80GB SXM5	80	FP16	65.5 est.	Yes	0.002
Gaudi 3 128GB	128	FP16	64.1 est.	Yes	0.005
RTX 5070 Ti 16GB	16	Q4	63.7 est.	Yes	0.085
RTX 4060 Ti 8GB	8	Q4	63.0 est.	No	0.158
RX 7600 8GB	8	Q4	63.0 est.	No	0.234
L40S 48GB	48	Q4	61.4 est.	Yes	0.008
RTX 4060 8GB	8	Q4	59.5 est.	No	0.199
Mac Studio M3 Ultra 96GB	96	Q4	58.2 est.	Yes	0.015
Mac Studio M2 Ultra 192GB	192	Q4	56.9 est.	Yes	0.008
RX 7900 XT 20GB	20	Q4	56.9 est.	Yes	0.063
RTX A6000 48GB	48	Q4	54.6 est.	Yes	0.012
RTX 3080 10GB	10	Q4	54.0 est.	Yes	0.077
RTX 4080 Super 16GB	16	Q4	52.3 est.	Yes	0.052
RTX 4070 Ti Super 16GB	16	Q4	47.8 est.	Yes	0.06
RTX 5070 12GB	12	Q4	47.8 est.	Yes	0.087
RX 9070 16GB	16	Q4	45.5 est.	Yes	0.083
RX 9070 XT 16GB	16	Q4	45.5 est.	Yes	0.076
RX 7800 XT 16GB	16	Q4	44.4 est.	Yes	0.089
RX 7900 XT 20GB	20	FP16	41.8 est.	No	0.046
RX 7900 GRE 16GB	16	Q4	41.0 est.	Yes	0.075
A100 80GB SXM4	80	FP16	39.9 est.	Yes	0.003
Arc A770 16GB	16	Q4	39.8 est.	Yes	0.114
RTX 4070 Ti 12GB	12	Q4	39.4 est.	Yes	0.049
MacBook Pro M4 Max 128GB	128	Q4	38.8 est.	Yes	0.009
RTX 4070 12GB	12	Q4	35.8 est.	Yes	0.06
RTX 4070 Super 12GB	12	Q4	35.8 est.	Yes	0.06
RTX 5090 32GB	32	FP16	35.0 est.	Yes	0.018
Arc B580 12GB	12	Q4	32.4 est.	Yes	0.13
RTX 5080 16GB	16	FP16	32.1 est.	No	0.032
RTX 5060 Ti 16GB	16	Q4	31.9 est.	Yes	0.074
RX 7700 XT 12GB	12	Q4	30.7 est.	Yes	0.068
RTX 5070 Ti 16GB	16	FP16	29.9 est.	No	0.04
MacBook Pro M3 Max 128GB	128	Q4	28.4 est.	Yes	0.007
Arc B570 10GB	10	Q4	27.0 est.	Yes	0.123
RTX 3060 12GB	12	Q4	25.6 est.	Yes	0.078
RTX 4080 Super 16GB	16	FP16	24.6 est.	No	0.025
T4 16GB	16	Q4	22.8 est.	Yes	0.011
RTX 4070 Ti Super 16GB	16	FP16	22.5 est.	No	0.028
RX 9070 16GB	16	FP16	21.4 est.	No	0.039
RX 9070 XT 16GB	16	FP16	21.4 est.	No	0.036
RX 7800 XT 16GB	16	FP16	20.8 est.	No	0.042
RTX 4060 Ti 16GB	16	Q4	20.5 est.	Yes	0.041
RX 7600 XT 16GB	16	Q4	20.5 est.	Yes	0.062
RTX 4090 24GB	24	FP16	19.7 est.	Yes	0.012
RTX 4090D 24GB (China)	24	FP16	19.7 est.	Yes	0.012
RTX 3090 Ti 24GB	24	FP16	19.7 est.	Yes	0.01
Mac mini M4 Pro 64GB	64	Q4	19.4 est.	Yes	0.009
RX 7900 GRE 16GB	16	FP16	19.2 est.	No	0.035
RX 7900 XTX 24GB	24	FP16	18.8 est.	Yes	0.019
RTX 6000 Ada 48GB	48	FP16	18.8 est.	Yes	0.003
Arc A770 16GB	16	FP16	18.7 est.	No	0.054
RTX 3090 24GB	24	FP16	18.3 est.	Yes	0.012
L40S 48GB	48	FP16	16.9 est.	Yes	0.002
Mac Studio M3 Ultra 96GB	96	FP16	16.0 est.	Yes	0.004
Mac Studio M2 Ultra 192GB	192	FP16	15.6 est.	Yes	0.002
RTX A6000 48GB	48	FP16	15.0 est.	Yes	0.003
RTX 5060 Ti 16GB	16	FP16	15.0 est.	No	0.035
RTX 5070 12GB	12	FP16	12.6 est.	No	0.023
MacBook Pro M4 Max 128GB	128	FP16	10.7 est.	Yes	0.002
T4 16GB	16	FP16	10.7 est.	No	0.005
RTX 4070 Ti 12GB	12	FP16	10.4 est.	No	0.013
RTX 3080 10GB	10	FP16	9.9 est.	No	0.014
RTX 4060 Ti 16GB	16	FP16	9.6 est.	No	0.019
RX 7600 XT 16GB	16	FP16	9.6 est.	No	0.029
RTX 4070 12GB	12	FP16	9.5 est.	No	0.016
RTX 4070 Super 12GB	12	FP16	9.5 est.	No	0.016
Arc B580 12GB	12	FP16	8.6 est.	No	0.035
Mac mini M4 32GB	32	Q4	8.5 est.	Yes	0.009
RX 7700 XT 12GB	12	FP16	8.1 est.	No	0.018
MacBook Pro M3 Max 128GB	128	FP16	7.8 est.	Yes	0.002
RTX 3060 12GB	12	FP16	6.8 est.	No	0.021
Mac mini M4 Pro 64GB	64	FP16	5.3 est.	Yes	0.002
Arc B570 10GB	10	FP16	5.0 est.	No	0.023
RTX 5060 8GB	8	FP16	3.7 est.	No	0.012
RTX 5060 Ti 8GB	8	FP16	3.7 est.	No	0.01
RTX 3070 8GB	8	FP16	3.7 est.	No	0.007
RTX 4060 Ti 8GB	8	FP16	2.4 est.	No	0.006
RX 7600 8GB	8	FP16	2.4 est.	No	0.009
RTX 4060 8GB	8	FP16	2.3 est.	No	0.008
Mac mini M4 32GB	32	FP16	2.3 est.	Yes	0.002

Fit checks

RTX 5090 32GB — fits · ≈127.4 tok/s Q4 est.
RTX 3090 Ti 24GB — fits · ≈71.7 tok/s Q4 est.
RTX 4090 24GB — fits · ≈71.7 tok/s Q4 est.
RTX 5080 16GB — fits · ≈68.3 tok/s Q4 est.
RTX 6000 Ada 48GB — fits · ≈68.3 tok/s Q4 est.
RX 7900 XTX 24GB — fits · ≈68.3 tok/s Q4 est.
RTX 3090 24GB — fits · ≈66.6 tok/s Q4 est.
RTX 5070 Ti 16GB — fits · ≈63.7 tok/s Q4 est.
L40S 48GB — fits · ≈61.4 tok/s Q4 est.
Mac Studio M3 Ultra 96GB — fits · ≈58.2 tok/s Q4 est.
Mac Studio M2 Ultra 192GB — fits · ≈56.9 tok/s Q4 est.
RX 7900 XT 20GB — fits · ≈56.9 tok/s Q4 est.

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

Same family: Qwen

Qwen3 Qwen3-8B · Qwen3-14B
Qwen3 2507 Qwen3-4B-Instruct-2507 · Qwen3-4B-Thinking-2507 · Qwen3-30B-A3B-Instruct-2507 · Qwen3-235B-A22B-Instruct-2507
Qwen2.5 Qwen2.5-7B-Instruct · Qwen2.5-14B-Instruct · Qwen2.5-32B-Instruct · Qwen2.5-72B-Instruct
Qwen2.5 Coder Qwen2.5-Coder-32B-Instruct
Qwen3 Next Qwen3-Next-80B-A3B-Instruct
Qwen3 Coder Qwen3-Coder-480B-A35B-Instruct · Qwen3-Coder-Next
Qwen3.5 Qwen3.5-2B · Qwen3.5-27B · Qwen3.5-35B-A3B · Qwen3.5-122B-A10B · Qwen3.5-397B-A17B
Qwen3.6 Qwen3.6-27B · Qwen3.6-35B-A3B

All Qwen models →