← AI Hashrate 中文

Qwen3.5-2B

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

Qwen3.5-2B (Alibaba) is a 2.0B parameters open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 2.5 GB VRAM for Q4 (weights 1.1 GB + KV 0.4 GB + 1 GB overhead) and 5.4 GB for FP16. About 50 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 2545.5 tok/s (estimated); MI300X 192GB ≈ 1686.4 tok/s (estimated); H200 141GB SXM5 ≈ 1527.3 tok/s (estimated); H100 80GB SXM5 ≈ 1065.9 tok/s (estimated); Gaudi 3 128GB ≈ 1042.4 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.

HardwareVRAMQuanttok/sFits?tok/s/$
B200 192GB SXM192Q42545.5 est.Yes0.064
MI300X 192GB192Q41686.4 est.Yes0.112
H200 141GB SXM5141Q41527.3 est.Yes0.044
H100 80GB SXM580Q41065.9 est.Yes0.036
Gaudi 3 128GB128Q41042.4 est.Yes0.083
B200 192GB SXM192FP16700.0 est.Yes0.018
A100 80GB SXM480Q4648.8 est.Yes0.043
RTX 5090 32GB32Q4570.2 est.Yes0.285
MI300X 192GB192FP16463.7 est.Yes0.031
H200 141GB SXM5141FP16420.0 est.Yes0.012
RTX 4090 24GB24Q4320.7 est.Yes0.201
RTX 4090D 24GB (China)24Q4320.7 est.Yes0.201
RTX 3090 Ti 24GB24Q4320.7 est.Yes0.16
RTX 5080 16GB16Q4305.5 est.Yes0.306
RX 7900 XTX 24GB24Q4305.5 est.Yes0.306
RTX 6000 Ada 48GB48Q4305.5 est.Yes0.045
RTX 3090 24GB24Q4297.8 est.Yes0.199
H100 80GB SXM580FP16293.1 est.Yes0.01
Gaudi 3 128GB128FP16286.6 est.Yes0.023
RTX 5070 Ti 16GB16Q4285.1 est.Yes0.381
L40S 48GB48Q4274.9 est.Yes0.034
Mac Studio M3 Ultra 96GB96Q4260.6 est.Yes0.065
Mac Studio M2 Ultra 192GB192Q4254.5 est.Yes0.036
RX 7900 XT 20GB20Q4254.5 est.Yes0.283
RTX A6000 48GB48Q4244.4 est.Yes0.054
RTX 3080 10GB10Q4241.8 est.Yes0.346
RTX 4080 Super 16GB16Q4234.2 est.Yes0.234
RTX 4070 Ti Super 16GB16Q4213.8 est.Yes0.268
RTX 5070 12GB12Q4213.8 est.Yes0.389
RX 9070 16GB16Q4203.6 est.Yes0.371
RX 9070 XT 16GB16Q4203.6 est.Yes0.34
RX 7800 XT 16GB16Q4198.5 est.Yes0.398
RX 7900 GRE 16GB16Q4183.3 est.Yes0.334
A100 80GB SXM480FP16178.4 est.Yes0.012
Arc A770 16GB16Q4178.2 est.Yes0.511
RTX 4070 Ti 12GB12Q4176.3 est.Yes0.221
MacBook Pro M4 Max 128GB128Q4173.7 est.Yes0.039
RTX 4070 12GB12Q4160.4 est.Yes0.268
RTX 4070 Super 12GB12Q4160.4 est.Yes0.268
RTX 5090 32GB32FP16156.8 est.Yes0.078
Arc B580 12GB12Q4145.1 est.Yes0.583
RTX 5060 8GB8Q4142.5 est.Yes0.477
RTX 5060 Ti 8GB8Q4142.5 est.Yes0.376
RTX 5060 Ti 16GB16Q4142.5 est.Yes0.332
RTX 3070 8GB8Q4142.5 est.Yes0.286
RX 7700 XT 12GB12Q4137.5 est.Yes0.306
MacBook Pro M3 Max 128GB128Q4127.3 est.Yes0.03
Arc B570 10GB10Q4120.9 est.Yes0.552
RTX 3060 12GB12Q4114.5 est.Yes0.348
T4 16GB16Q4101.8 est.Yes0.051
RTX 4060 Ti 8GB8Q491.6 est.Yes0.23
RTX 4060 Ti 16GB16Q491.6 est.Yes0.184
RX 7600 8GB8Q491.6 est.Yes0.341
RX 7600 XT 16GB16Q491.6 est.Yes0.278
RTX 4090 24GB24FP1688.2 est.Yes0.055
RTX 4090D 24GB (China)24FP1688.2 est.Yes0.055
RTX 3090 Ti 24GB24FP1688.2 est.Yes0.044
Mac mini M4 Pro 64GB64Q486.9 est.Yes0.04
RTX 4060 8GB8Q486.5 est.Yes0.289
RTX 5080 16GB16FP1684.0 est.Yes0.084
RX 7900 XTX 24GB24FP1684.0 est.Yes0.084
RTX 6000 Ada 48GB48FP1684.0 est.Yes0.012
RTX 3090 24GB24FP1681.9 est.Yes0.055
RTX 5070 Ti 16GB16FP1678.4 est.Yes0.105
L40S 48GB48FP1675.6 est.Yes0.009
Mac Studio M3 Ultra 96GB96FP1671.7 est.Yes0.018
Mac Studio M2 Ultra 192GB192FP1670.0 est.Yes0.01
RX 7900 XT 20GB20FP1670.0 est.Yes0.078
RTX A6000 48GB48FP1667.2 est.Yes0.015
RTX 3080 10GB10FP1666.5 est.Yes0.095
RTX 4080 Super 16GB16FP1664.4 est.Yes0.064
RTX 4070 Ti Super 16GB16FP1658.8 est.Yes0.074
RTX 5070 12GB12FP1658.8 est.Yes0.107
RX 9070 16GB16FP1656.0 est.Yes0.102
RX 9070 XT 16GB16FP1656.0 est.Yes0.093
RX 7800 XT 16GB16FP1654.6 est.Yes0.109
RX 7900 GRE 16GB16FP1650.4 est.Yes0.092
Arc A770 16GB16FP1649.0 est.Yes0.14
RTX 4070 Ti 12GB12FP1648.5 est.Yes0.061
MacBook Pro M4 Max 128GB128FP1647.8 est.Yes0.011
RTX 4070 12GB12FP1644.1 est.Yes0.074
RTX 4070 Super 12GB12FP1644.1 est.Yes0.074
Arc B580 12GB12FP1639.9 est.Yes0.16
RTX 5060 8GB8FP1639.2 est.Yes0.131
RTX 5060 Ti 8GB8FP1639.2 est.Yes0.103
RTX 5060 Ti 16GB16FP1639.2 est.Yes0.091
RTX 3070 8GB8FP1639.2 est.Yes0.079
Mac mini M4 32GB32Q438.2 est.Yes0.038
RX 7700 XT 12GB12FP1637.8 est.Yes0.084
MacBook Pro M3 Max 128GB128FP1635.0 est.Yes0.008
Arc B570 10GB10FP1633.2 est.Yes0.152
RTX 3060 12GB12FP1631.5 est.Yes0.096
T4 16GB16FP1628.0 est.Yes0.014
RTX 4060 Ti 8GB8FP1625.2 est.Yes0.063
RTX 4060 Ti 16GB16FP1625.2 est.Yes0.051
RX 7600 8GB8FP1625.2 est.Yes0.094
RX 7600 XT 16GB16FP1625.2 est.Yes0.077
Mac mini M4 Pro 64GB64FP1623.9 est.Yes0.011
RTX 4060 8GB8FP1623.8 est.Yes0.08
Mac mini M4 32GB32FP1610.5 est.Yes0.011

Fit checks

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 Qwen3.5-9B · 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 →