Gemma 2 models
11 Gemma 2 models from Google on Hugging Face, from 2.5B to 27.2B parameters, published by Google in BF16 and F32, with a 8K-token context. The smallest official model, gemma-2b, needs about 5.6 GB of VRAM at its published precision; the cheapest live fit is RTX 3070 at $0.088/hr.
Part of the Gemma series · Previous generation: Gemma 1 · Next generation: Gemma 3
Pick a size
One row per official Gemma 2 size: the VRAM it needs at each precision, the cheapest GPU that holds it today, and the KV cache for a 32K-token context.
| Model | Parameters | Native VRAM | FP8 VRAM | INT4 VRAM | Live GPU fit (native) | Est. $/hr | KV cache at 32K |
|---|---|---|---|---|---|---|---|
| gemma-2b-it | 2.5B | 5.6 GB | 2.8 GB | 1.4 GB | RTX 3070 | $0.088/hr | 0.56 GB |
| gemma-2-2b-it | 2.6B | 5.8 GB | 2.9 GB | 1.5 GB | RTX 3070 | $0.088/hr | 1.8 GB |
| gemma-2-9b-it | 9.2B | 20.7 GB | 10.3 GB | 5.2 GB | RTX A5000 | $0.176/hr | 5.9 GB |
| gemma-2-27b-it | 27.2B | 60.9 GB | 30.4 GB | 15.2 GB | A100 | $1.31/hr | 6.5 GB |
VRAM is weights times a flat 1.2 overhead; the KV cache is a separate per-model figure at 16-bit, shown where the architecture is published. See the methodology.
Official models (8)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| gemma-2b | 2.5B | BF16 | 5.6 GB | RTX 3070 | $0.088/hr |
| gemma-2b-it | 2.5B | BF16 | 5.6 GB | RTX 3070 | $0.088/hr |
| gemma-2-2b | 2.6B | F32 | 11.7 GB | V100 | $0.110/hr |
| gemma-2-2b-it | 2.6B | BF16 | 5.8 GB | RTX 3070 | $0.088/hr |
| gemma-2-9b | 9.2B | F32 | 41.3 GB | RTX A6000 | $0.363/hr |
| gemma-2-9b-it | 9.2B | BF16 | 20.7 GB | RTX A5000 | $0.176/hr |
| gemma-2-27b | 27.2B | F32 | 122 GB | AMD MI300X | $2.63/hr |
| gemma-2-27b-it | 27.2B | BF16 | 60.9 GB | A100 | $1.31/hr |
Fine-tunes and community models (3)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| gemma-2-2b | 2.6B | BF16 | 5.8 GB | RTX 3070 | $0.088/hr |
| gemma-2-9b | 9.2B | BF16 | 20.7 GB | RTX A5000 | $0.176/hr |
| gemma-2-27b | 27.2B | BF16 | 60.9 GB | A100 | $1.31/hr |
VRAM is for the precision the model is published in: the weights times a flat 1.2 overhead, with the KV cache not included. See the methodology. The fit shown is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does.
Run and fine-tune Gemma 2
- How to run Gemma 2: engine support, chat template and context flags, with a serving command sized per model.
- How to fine-tune Gemma 2: LoRA and QLoRA memory per model, and the license limits on derivatives.