Gemma 4 12B locally: the VRAM, the GPU, and one Ollama gotcha

Google's Gemma 4 12B is the multimodal open model that fits a mid-range card — if you pull the right Ollama tag. The honest VRAM math, the cheapest… Aliteq Lena Fischer · AI & Local Compute Editor

Gemma 4 12B is Google's mid-size multimodal open model (11.95B parameters, 256K context, text + image + audio). At Q4_K_M our VRAM engine computes about 7.8 GB, so it runs on a 12GB card like the…

11.95B params, dense, multimodal, Apache-2.0 (per the model card) — read on Hugging Face.

Q4_K_M ~7.8 GB, Q8_0 ~12.9 GB, BF16 ~23 GB (computed by our VRAM engine, 16K context).

Fits a 12GB card at Q4; pull it as gemma4:12b, not the bare tag.

Read the full story

Aliteq Gemma 4 12B locally: the VRAM, the GPU, and one Ollama gotcha Read the full story on Aliteq