← AI Hashrate 中文

gemma-3-27b-it

google · 27.4B · ctx NoneK · gemma · HuggingFace

gemma-3-27b-it (google) is a 27.4B parameters open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 21.55 GB VRAM for Q4 (weights 15.07 GB + KV 5.48 GB + 1 GB overhead) and 61.28 GB for FP16. About 28 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 185.8 tok/s (estimated); MI300X 192GB ≈ 123.1 tok/s (estimated); H200 141GB SXM5 ≈ 111.5 tok/s (estimated); H100 80GB SXM5 ≈ 77.8 tok/s (estimated); Gaudi 3 128GB ≈ 76.1 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.

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

HardwareVRAMQuanttok/sFits?tok/s/$
B200 192GB SXM192Q4185.8 est.Yes0.005
MI300X 192GB192Q4123.1 est.Yes0.008
H200 141GB SXM5141Q4111.5 est.Yes0.003
H100 80GB SXM580Q477.8 est.Yes0.003
Gaudi 3 128GB128Q476.1 est.Yes0.006
B200 192GB SXM192FP1651.1 est.Yes0.001
A100 80GB SXM480Q447.4 est.Yes0.003
RX 7900 XT 20GB20Q445.6 est.No0.051
RTX 5090 32GB32Q441.6 est.Yes0.021
RTX 5080 16GB16Q435.0 est.No0.035
MI300X 192GB192FP1633.9 est.Yes0.002
RTX 5070 Ti 16GB16Q432.7 est.No0.044
H200 141GB SXM5141FP1630.7 est.Yes0.001
RTX 4080 Super 16GB16Q426.8 est.No0.027
RTX 4070 Ti Super 16GB16Q424.5 est.No0.031
RTX 4090 24GB24Q423.4 est.Yes0.015
RTX 4090D 24GB (China)24Q423.4 est.Yes0.015
RTX 3090 Ti 24GB24Q423.4 est.Yes0.012
RX 9070 16GB16Q423.3 est.No0.042
RX 9070 XT 16GB16Q423.3 est.No0.039
RX 7800 XT 16GB16Q422.8 est.No0.046
RX 7900 XTX 24GB24Q422.3 est.Yes0.022
RTX 6000 Ada 48GB48Q422.3 est.Yes0.003
RTX 3090 24GB24Q421.7 est.Yes0.014
H100 80GB SXM580FP1621.4 est.Yes0.001
RX 7900 GRE 16GB16Q421.0 est.No0.038
Gaudi 3 128GB128FP1620.9 est.Yes0.002
Arc A770 16GB16Q420.4 est.No0.058
L40S 48GB48Q420.1 est.Yes0.003
Mac Studio M3 Ultra 96GB96Q419.0 est.Yes0.005
Mac Studio M2 Ultra 192GB192Q418.6 est.Yes0.003
RTX A6000 48GB48Q417.8 est.Yes0.004
RTX 5060 Ti 16GB16Q416.3 est.No0.038
Arc Pro B70 32GB32Q414.1 est.Yes0.014
RTX 5070 12GB12Q413.8 est.No0.025
Titan V 12GB12Q413.4 est.No0.004
A100 80GB SXM480FP1613.0 est.Yes0.001
MacBook Pro M4 Max 128GB128Q412.7 est.Yes0.003
T4 16GB16Q411.7 est.No0.006
DGX Spark (GH200 480GB)480Q411.6 est.Yes0.004
RTX 4070 Ti 12GB12Q411.4 est.No0.014
Titan Xp 12GB12Q411.2 est.No0.009
RTX 3080 10GB10Q410.8 est.No0.015
RTX 6000 Ada 48GB48FP1610.7 est.No0.002
RTX 4060 Ti 16GB16Q410.5 est.No0.021
RX 7600 XT 16GB16Q410.5 est.No0.032
RTX 4070 12GB12Q410.3 est.No0.017
RTX 4070 Super 12GB12Q410.3 est.No0.017
Titan X Pascal 12GB12Q49.8 est.No0.008
L40S 48GB48FP169.7 est.No0.001
MacBook Pro M1 Max 32GB32Q49.3 est.Yes0.003
MacBook Pro M1 Max 64GB64Q49.3 est.Yes0.002
MacBook Pro M3 Max 128GB128Q49.3 est.Yes0.002
Arc B580 12GB12Q49.3 est.No0.037
RTX 5090 32GB32FP168.9 est.No0.004
RX 7700 XT 12GB12Q48.9 est.No0.02
RTX A6000 48GB48FP168.6 est.No0.002
RTX 3060 12GB12Q47.4 est.No0.022
Mac mini M4 Pro 64GB64Q46.3 est.Yes0.003
Arc B570 10GB10Q45.4 est.No0.025
Mac Studio M3 Ultra 96GB96FP165.2 est.Yes0.001
Mac Studio M2 Ultra 192GB192FP165.1 est.Yes0.001
Ryzen AI Max+ 395 128GB96Q45.0 est.Yes0.003
Ryzen AI Max+ 395 64GB48Q45.0 est.Yes0.003
Ryzen AI Max 390 64GB48Q45.0 est.Yes0.004
RTX 5060 8GB8Q44.1 est.No0.014
RTX 5060 Ti 8GB8Q44.1 est.No0.011
RTX 3070 8GB8Q44.1 est.No0.008
MacBook Pro M4 Max 128GB128FP163.5 est.Yes0.001
DGX Spark (GH200 480GB)480FP163.2 est.Yes0.001
Arc Pro B70 32GB32FP163.0 est.No0.003
RTX 4090 24GB24FP162.8 est.No0.002
RTX 4090D 24GB (China)24FP162.8 est.No0.002
RTX 3090 Ti 24GB24FP162.8 est.No0.001
Mac mini M4 32GB32Q42.8 est.Yes0.003
RX 7900 XTX 24GB24FP162.7 est.No0.003
RTX 3090 24GB24FP162.6 est.No0.002
MacBook Pro M1 Max 64GB64FP162.6 est.Yes0.001
MacBook Pro M3 Max 128GB128FP162.6 est.Yes0.001
RTX 4060 Ti 8GB8Q42.6 est.No0.007
RX 7600 8GB8Q42.6 est.No0.01
RTX 4060 8GB8Q42.5 est.No0.008
Ryzen AI Max+ 395 64GB48FP162.4 est.No0.002
Ryzen AI Max 390 64GB48FP162.4 est.No0.002
MacBook Pro M1 Max 32GB32FP162.0 est.No0.001
Tesla P4 8GB8Q41.7 est.No0.021
Mac mini M4 Pro 64GB64FP161.7 est.Yes0.001
RX 7900 XT 20GB20FP161.6 est.No0.002
Ryzen AI Max+ 395 128GB96FP161.4 est.Yes0.001
RTX 5080 16GB16FP161.2 est.No0.001
RTX 5070 Ti 16GB16FP161.1 est.No0.001
RTX 4080 Super 16GB16FP160.9 est.No0.001
RTX 4070 Ti Super 16GB16FP160.8 est.No0.001
RX 7800 XT 16GB16FP160.8 est.No0.002
RX 9070 16GB16FP160.8 est.No0.001
RX 9070 XT 16GB16FP160.8 est.No0.001
RX 7900 GRE 16GB16FP160.7 est.No0.001
Arc A770 16GB16FP160.7 est.No0.002
RTX 5060 Ti 16GB16FP160.6 est.No0.001
Mac mini M4 32GB32FP160.6 est.No0.001
T4 16GB16FP160.4 est.No0.0
RTX 4060 Ti 16GB16FP160.4 est.No0.001
RX 7600 XT 16GB16FP160.4 est.No0.001

Fit checks

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