Qwen3.5-27B — which GPUs fit

Alibaba · 27.0B · ctx 256K · apache-2.0 · HuggingFace

Qwen3.5-27B (Alibaba) is a 27.0B parameters open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 21.25 GB VRAM for Q4 (weights 14.85 GB + KV 5.4 GB + 1 GB overhead) and 60.4 GB for FP16. About 21 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 188.6 tok/s (estimated); MI300X 192GB ≈ 124.9 tok/s (estimated); H200 141GB SXM5 ≈ 113.1 tok/s (estimated); H100 80GB SXM5 ≈ 79.0 tok/s (estimated); Gaudi 3 128GB ≈ 77.2 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	188.6 est.	Yes	0.005
MI300X 192GB	192	Q4	124.9 est.	Yes	0.008
H200 141GB SXM5	141	Q4	113.1 est.	Yes	0.003
H100 80GB SXM5	80	Q4	79.0 est.	Yes	0.003
Gaudi 3 128GB	128	Q4	77.2 est.	Yes	0.006
B200 192GB SXM	192	FP16	51.9 est.	Yes	0.001
A100 80GB SXM4	80	Q4	48.1 est.	Yes	0.003
RX 7900 XT 20GB	20	Q4	47.7 est.	No	0.053
RTX 5090 32GB	32	Q4	42.2 est.	Yes	0.021
RTX 5080 16GB	16	Q4	36.6 est.	No	0.037
MI300X 192GB	192	FP16	34.4 est.	Yes	0.002
RTX 5070 Ti 16GB	16	Q4	34.2 est.	No	0.046
H200 141GB SXM5	141	FP16	31.1 est.	Yes	0.001
RTX 4080 Super 16GB	16	Q4	28.1 est.	No	0.028
RTX 4070 Ti Super 16GB	16	Q4	25.7 est.	No	0.032
RX 9070 16GB	16	Q4	24.4 est.	No	0.044
RX 9070 XT 16GB	16	Q4	24.4 est.	No	0.041
RTX 4090 24GB	24	Q4	23.8 est.	Yes	0.015
RTX 4090D 24GB (China)	24	Q4	23.8 est.	Yes	0.015
RTX 3090 Ti 24GB	24	Q4	23.8 est.	Yes	0.012
RX 7800 XT 16GB	16	Q4	23.8 est.	No	0.048
RX 7900 XTX 24GB	24	Q4	22.6 est.	Yes	0.023
RTX 6000 Ada 48GB	48	Q4	22.6 est.	Yes	0.003
RTX 3090 24GB	24	Q4	22.1 est.	Yes	0.015
RX 7900 GRE 16GB	16	Q4	22.0 est.	No	0.04
H100 80GB SXM5	80	FP16	21.7 est.	Yes	0.001
Arc A770 16GB	16	Q4	21.4 est.	No	0.061
Gaudi 3 128GB	128	FP16	21.2 est.	Yes	0.002
L40S 48GB	48	Q4	20.4 est.	Yes	0.003
Mac Studio M3 Ultra 96GB	96	Q4	19.3 est.	Yes	0.005
Mac Studio M2 Ultra 192GB	192	Q4	18.9 est.	Yes	0.003
RTX A6000 48GB	48	Q4	18.1 est.	Yes	0.004
RTX 5060 Ti 16GB	16	Q4	17.1 est.	No	0.04
RTX 5070 12GB	12	Q4	14.4 est.	No	0.026
A100 80GB SXM4	80	FP16	13.2 est.	Yes	0.001
MacBook Pro M4 Max 128GB	128	Q4	12.9 est.	Yes	0.003
T4 16GB	16	Q4	12.2 est.	No	0.006
RTX 4070 Ti 12GB	12	Q4	11.9 est.	No	0.015
RTX 3080 10GB	10	Q4	11.3 est.	No	0.016
RTX 6000 Ada 48GB	48	FP16	11.2 est.	No	0.002
RTX 4060 Ti 16GB	16	Q4	11.0 est.	No	0.022
RX 7600 XT 16GB	16	Q4	11.0 est.	No	0.033
RTX 4070 12GB	12	Q4	10.8 est.	No	0.018
RTX 4070 Super 12GB	12	Q4	10.8 est.	No	0.018
L40S 48GB	48	FP16	10.1 est.	No	0.001
Arc B580 12GB	12	Q4	9.8 est.	No	0.039
MacBook Pro M3 Max 128GB	128	Q4	9.4 est.	Yes	0.002
RTX 5090 32GB	32	FP16	9.3 est.	No	0.005
RX 7700 XT 12GB	12	Q4	9.3 est.	No	0.021
RTX A6000 48GB	48	FP16	9.0 est.	No	0.002
RTX 3060 12GB	12	Q4	7.7 est.	No	0.023
Mac mini M4 Pro 64GB	64	Q4	6.4 est.	Yes	0.003
Arc B570 10GB	10	Q4	5.7 est.	No	0.026
Mac Studio M3 Ultra 96GB	96	FP16	5.3 est.	Yes	0.001
Mac Studio M2 Ultra 192GB	192	FP16	5.2 est.	Yes	0.001
RTX 5060 8GB	8	Q4	4.3 est.	No	0.014
RTX 5060 Ti 8GB	8	Q4	4.3 est.	No	0.011
RTX 3070 8GB	8	Q4	4.3 est.	No	0.009
MacBook Pro M4 Max 128GB	128	FP16	3.5 est.	Yes	0.001
RTX 4090 24GB	24	FP16	2.9 est.	No	0.002
RTX 4090D 24GB (China)	24	FP16	2.9 est.	No	0.002
RTX 3090 Ti 24GB	24	FP16	2.9 est.	No	0.001
RX 7900 XTX 24GB	24	FP16	2.8 est.	No	0.003
Mac mini M4 32GB	32	Q4	2.8 est.	Yes	0.003
RTX 3090 24GB	24	FP16	2.7 est.	No	0.002
RTX 4060 Ti 8GB	8	Q4	2.7 est.	No	0.007
RX 7600 8GB	8	Q4	2.7 est.	No	0.01
MacBook Pro M3 Max 128GB	128	FP16	2.6 est.	Yes	0.001
RTX 4060 8GB	8	Q4	2.6 est.	No	0.009
Mac mini M4 Pro 64GB	64	FP16	1.8 est.	Yes	0.001
RX 7900 XT 20GB	20	FP16	1.6 est.	No	0.002
RTX 5080 16GB	16	FP16	1.2 est.	No	0.001
RTX 5070 Ti 16GB	16	FP16	1.2 est.	No	0.002
RTX 4080 Super 16GB	16	FP16	1.0 est.	No	0.001
RTX 4070 Ti Super 16GB	16	FP16	0.9 est.	No	0.001
RX 7800 XT 16GB	16	FP16	0.8 est.	No	0.002
RX 9070 16GB	16	FP16	0.8 est.	No	0.001
RX 9070 XT 16GB	16	FP16	0.8 est.	No	0.001
RX 7900 GRE 16GB	16	FP16	0.7 est.	No	0.001
Arc A770 16GB	16	FP16	0.7 est.	No	0.002
RTX 5060 Ti 16GB	16	FP16	0.6 est.	No	0.001
Mac mini M4 32GB	32	FP16	0.6 est.	No	0.001
T4 16GB	16	FP16	0.4 est.	No	0.0
RTX 4060 Ti 16GB	16	FP16	0.4 est.	No	0.001
RX 7600 XT 16GB	16	FP16	0.4 est.	No	0.001

Fit checks

RTX 5090 32GB — fits · ≈42.2 tok/s Q4 est.
RTX 3090 Ti 24GB — fits · ≈23.8 tok/s Q4 est.
RTX 4090 24GB — fits · ≈23.8 tok/s Q4 est.
RTX 5080 16GB — does not fit · ≈36.6 tok/s Q4 est.
RTX 6000 Ada 48GB — fits · ≈22.6 tok/s Q4 est.
RX 7900 XTX 24GB — fits · ≈22.6 tok/s Q4 est.
RTX 3090 24GB — fits · ≈22.1 tok/s Q4 est.
RTX 5070 Ti 16GB — does not fit · ≈34.2 tok/s Q4 est.
L40S 48GB — fits · ≈20.4 tok/s Q4 est.
Mac Studio M3 Ultra 96GB — fits · ≈19.3 tok/s Q4 est.
Mac Studio M2 Ultra 192GB — fits · ≈18.9 tok/s Q4 est.
RX 7900 XT 20GB — does not fit · ≈47.7 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 · Qwen3.5-9B · Qwen3.5-35B-A3B · Qwen3.5-122B-A10B · Qwen3.5-397B-A17B
Qwen3.6 Qwen3.6-27B · Qwen3.6-35B-A3B

All Qwen models →