Can the NVIDIA RTX 3090 run Gemma 3 12B?
Yes
QuantQ8_0Near-lossless
Memory16.8 GBof 24 GB
Speed~43tokens/s
Yes. The NVIDIA RTX 3090 can run Gemma 3 12B fully in VRAM at Q8_0 (near-lossless), using about 16.8 GB of the 24 GB available, at roughly 43 tokens/second.
Try other context lengths in the calculator →Memory needed by quantization
At 8,192 tokens of context.
| Quant | Quality | Memory | Fits? | Speed |
|---|---|---|---|---|
| FP16 | Full quality | 28.2 GB | Too big | — |
| Q8_0 | Near-lossless | 16.8 GB | Fits | ~43 tok/s |
| Q6_K | Excellent | 13.8 GB | Fits | ~56 tok/s |
| Q5_K_M | Very good | 12.5 GB | Fits | ~65 tok/s |
| Q4_K_M | Good, most popular | 11.2 GB | Fits | ~76 tok/s |
| Q3_K_M | Noticeably degraded | 9.8 GB | Fits | ~94 tok/s |
| Q2_K | Heavily degraded | 8.4 GB | Fits | ~123 tok/s |
How to run Gemma 3 12B on the NVIDIA RTX 3090
With Ollama installed, one command downloads and starts it:
ollama run gemma3:12bOllama's default download is Q4_K_M (about 11.2 GB here). You have room for Q8_0, which is near-lossless: pick that GGUF in LM Studio or from Hugging Face.
GPUs that run Gemma 3 12B comfortably
- NVIDIA RTX 3060 12GB
- NVIDIA RTX 4070 (12 GB)
- NVIDIA RTX 5070 (12 GB)
- NVIDIA RTX 4060 Ti 16GB
- NVIDIA RTX 5060 Ti 16GB
Models that run well on the NVIDIA RTX 3090
Run it in the cloud
Don't want to buy hardware? Run Gemma 3 12B on a rented GPU today.
Rent a cloud GPU by the hour. No upfront cost, cancel any time.