How much VRAM does Gemma 4 12B need?
Gemma 4 12B needs 6.8 GB for its weights at 4-bit, 8.2 GB with an 8K context and 8.4 GB with 32K. At 8-bit, 13.3 GB.
- Parameters
- 12B
- Longest context
- 256K
- Released
- May 2026
- License
- Apache 2.0
- Strongest at
- Chat, Coding, Docs
Memory at each precision
| Precision | Weights | With 8K | With 32K | With 128K |
|---|---|---|---|---|
| 16-bit | 22.3 GB | 23.7 GB | 23.9 GB | 24.7 GB |
| 8-bit | 11.8 GB | 13.3 GB | 13.5 GB | 14.2 GB |
| 6-bit | 9.1 GB | 10.6 GB | 10.8 GB | 11.5 GB |
| 5-bit | 7.9 GB | 9.4 GB | 9.6 GB | 10.3 GB |
| 4-bit | 6.8 GB | 8.2 GB | 8.4 GB | 9.1 GB |
| 3-bit | 5.4 GB | 6.9 GB | 7.1 GB | 7.8 GB |
| 2-bit | 4.2 GB | 5.6 GB | 5.8 GB | 6.6 GB |
| 1.6-bit | 2.3 GB | 3.7 GB | 3.9 GB | 4.7 GB |
One conversation, 16-bit context cache, runtime overhead included.
What runs it
| Machine | Best fit | Context | Speed | First reply |
|---|---|---|---|---|
| RTX 3060 12 GB | 6-bit | 112K | 25 tok/s | 1.9 s |
| RTX 4060 Ti 16 GB | 8-bit | 248K | 16 tok/s | 1.1 s |
| RTX 3090 | 8-bit | 256K | 48 tok/s | 0.7 s |
| RTX 4090 | 8-bit | 256K | 51 tok/s | 0.3 s |
| RTX 5090 | 8-bit | 256K | 86 tok/s | 0.2 s |
| 2× RTX 3090 | 8-bit | 256K | 48 tok/s | 0.4 s |
| RTX PRO 6000 Blackwell | 8-bit | 256K | 86 tok/s | 0.2 s |
| M4 Max 128 GB | 8-bit | 256K | 25 tok/s | 3.9 s |
| M5 Max 128 GB | 8-bit | 256K | 28 tok/s | 1.1 s |
| Ryzen AI Max+ 395 128 GB | 8-bit | 256K | 12 tok/s | 2.2 s |
| DGX Spark 128 GB | 8-bit | 256K | 15 tok/s | 0.8 s |
| M3 Ultra 256 GB | 8-bit | 256K | 36 tok/s | 2.2 s |
| M5 Ultra 512 GB | 8-bit | 256K | 50 tok/s | 0.7 s |
| H100 SXM | 8-bit | 256K | 146 tok/s | 0.2 s |
| DGX Station GB300 | 8-bit | 256K | 250 tok/s | 0.1 s |
| 8× H200 NVL | 8-bit | 256K | 191 tok/s | 0.1 s |
Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.
How to run it
- Ollama
ollama run gemma4:12b- llama.cpp
llama-server -hf unsloth/gemma-4-12b-it-GGUF- vLLM
vllm serve google/gemma-4-12B-it- MLX (Mac)
mlx_lm.chat --model mlx-community/gemma-4-12B-it-4bit
Questions
How much VRAM does Gemma 4 12B need?
Gemma 4 12B needs 6.8 GB for its weights at 4-bit, 8.2 GB with an 8K context and 8.4 GB with 32K. At 8-bit, 13.3 GB.
Can Gemma 4 12B run on an RTX 4090?
Yes. At 8-bit it holds 256K of context and writes about 51 tokens a second.