Gemma 4 12B

Google · google/gemma-4-12b

ProviderGoogle
Familygemma
Typevlm
Statusactive
Released2026-05-23
Updated2026-07-15
Parameters12B
Open weightsyes

What it fits on

Ordered by predicted speed. Bandwidth sets decode rate; memory decides whether it runs at all.

Every figure here is computed, not measured. Fit is weights at each quantisation against device memory, with a 25% allowance for the KV cache, activations and the OS. Decode rate is the memory-bandwidth roofline at 70% efficiency. Nobody has run this model on these devices.
DeviceMemoryBandwidthBest qualityWeightstok/sFastest
Cerebras WSE-344.0 GB21000000.0 GB/sbf1623.92 GB~614562.4~2458249.4 q4
NVIDIA GB200 Grace Blackwell Superchip372.0 GB16000.0 GB/sbf1623.92 GB~468.2~1873.0 q4
AMD Instinct MI355X288.0 GB8000.0 GB/sbf1623.92 GB~234.1~936.5 q4
NVIDIA B200180.0 GB7700.0 GB/sbf1623.92 GB~225.3~901.4 q4
Google TPU7x (Ironwood)192.0 GB7380.0 GB/sbf1623.92 GB~216.0~863.9 q4
AMD Instinct MI325X256.0 GB6000.0 GB/sbf1623.92 GB~175.6~702.4 q4
AMD Instinct MI300X192.0 GB5300.0 GB/sbf1623.92 GB~155.1~620.4 q4
NVIDIA H200 SXM141.0 GB4800.0 GB/sbf1623.92 GB~140.5~561.9 q4
NVIDIA H100 NVL94.0 GB3900.0 GB/sbf1623.92 GB~114.1~456.5 q4
NVIDIA H100 SXM80.0 GB3350.0 GB/sbf1623.92 GB~98.0~392.1 q4
AMD Instinct MI250X128.0 GB3200.0 GB/sbf1623.92 GB~93.6~374.6 q4
Google TPU v5p95.0 GB2765.0 GB/sbf1623.92 GB~80.9~323.7 q4
NVIDIA A100 80GB SXM80.0 GB2039.0 GB/sbf1623.92 GB~59.7~238.7 q4
NVIDIA H100 PCIe80.0 GB2000.0 GB/sbf1623.92 GB~58.5~234.1 q4
NVIDIA GeForce RTX 509032.0 GB1792.0 GB/sbf1623.92 GB~52.4~209.8 q4
Google TPU v6e (Trillium)32.0 GB1638.0 GB/sbf1623.92 GB~47.9~191.7 q4
AMD Instinct MI21064.0 GB1600.0 GB/sbf1623.92 GB~46.8~187.3 q4
NVIDIA A100 40GB SXM40.0 GB1555.0 GB/sbf1623.92 GB~45.5~182.0 q4
Google TPU v432.0 GB1200.0 GB/sbf1623.92 GB~35.1~140.5 q4
NVIDIA GeForce RTX 409024.0 GB1008.0 GB/sfp811.96 GB~59.0~118.0 q4
AMD Radeon RX 7900 XTX24.0 GB960.0 GB/sfp811.96 GB~56.2~112.4 q4
NVIDIA GeForce RTX 508016.0 GB960.0 GB/sfp811.96 GB~56.2~112.4 q4
NVIDIA GeForce RTX 309024.0 GB936.0 GB/sfp811.96 GB~54.8~109.6 q4
NVIDIA L40S48.0 GB864.0 GB/sbf1623.92 GB~25.3~101.1 q4
Apple M3 Ultra512.0 GB819.0 GB/sbf1623.92 GB~24.0~95.9 q4
AMD Radeon RX 7900 XT20.0 GB800.0 GB/sfp811.96 GB~46.8~93.6 q4
Apple M2 Ultra192.0 GB800.0 GB/sbf1623.92 GB~23.4~93.6 q4
Google TPU v5e16.0 GB800.0 GB/sfp811.96 GB~46.8~93.6 q4
NVIDIA GeForce RTX 4080 SUPER16.0 GB736.0 GB/sfp811.96 GB~43.1~86.2 q4
NVIDIA GeForce RTX 4070 Ti SUPER16.0 GB672.0 GB/sfp811.96 GB~39.3~78.7 q4
AMD Radeon RX 9070 XT16.0 GB640.0 GB/sfp811.96 GB~37.5~74.9 q4
Apple M4 Max128.0 GB546.0 GB/sbf1623.92 GB~16.0~63.9 q4
Apple M1 Max64.0 GB400.0 GB/sbf1623.92 GB~11.7~46.8 q4
Apple M2 Max96.0 GB400.0 GB/sbf1623.92 GB~11.7~46.8 q4
Apple M3 Max128.0 GB400.0 GB/sbf1623.92 GB~11.7~46.8 q4
NVIDIA GeForce RTX 3060 12GB12.0 GB360.0 GB/sq68.97 GB~28.1~42.1 q4
NVIDIA L424.0 GB300.0 GB/sfp811.96 GB~17.6~35.1 q4
NVIDIA GeForce RTX 4060 Ti 16GB16.0 GB288.0 GB/sfp811.96 GB~16.9~33.7 q4
Apple M4 Pro64.0 GB273.0 GB/sbf1623.92 GB~8.0~32.0 q4
Apple M432.0 GB120.0 GB/sbf1623.92 GB~3.5~14.0 q4

Where it runs

Taken from the card's availability section.

PlatformId on that platform
Hugging Facegoogle/gemma-4-12B

Reported benchmark scores

This card reports no benchmark scores yet.

BenchmarkCatalogue standingScore

Data

This card as JSON · See it in the graph · Edit on GitHub