Models / Gemma
Can an RTX 3060 12GB run Gemma 4 12B?
Dense Gemma 4 at 11.95B parameters, with a 256k context on the model card.
- Stored parameters
- 12B
- Active parameters
- Dense
- Q4 total (estimate)
- 8.0 GB
- License
- See model card
Model card. Google's Gemma 4 card lists the 12B unified model at 11.95B parameters.
Compatibility
| Hardware | Q4 | Q8 | FP16 |
|---|---|---|---|
| Ryzen AI Max+ 395 64GB AMD · 64 GB unified | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run | |
| RTX 4090 NVIDIA · 24 GB VRAM | 13.4 GB Fits estimate No measured run | 24.6 GB Over estimate No measured run | |
| RTX 5070 Ti NVIDIA · 16 GB VRAM | 13.4 GB Fits estimate No measured run | 24.6 GB Over estimate No measured run | |
| M4 16GB Apple · 16 GB unified | 13.4 GB Tight estimate No measured run | 24.6 GB Over estimate No measured run | |
| RTX 2060 Super NVIDIA · 8 GB VRAM | 13.4 GB Over estimate No measured run | 24.6 GB Over estimate No measured run | |
| M3 Ultra 512GB Apple · 512 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| 2× NVIDIA GB10 NVIDIA · 256 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| M3 Ultra 256GB Apple · 256 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| M5 Ultra 256GB Apple · 256 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| M2 Ultra 192GB Apple · 192 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| NVIDIA GB10 NVIDIA · 128 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| Ryzen AI Max+ 395 128GB AMD · 128 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| M1 Ultra 128GB Apple · 128 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| M5 Max 128GB Apple · 128 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| 4× RTX 3090 NVIDIA · 96 GB VRAM | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| RTX Pro 6000 Blackwell NVIDIA · 96 GB VRAM | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| M5 Ultra 96GB Apple · 96 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| 2× RTX 3090 NVIDIA · 48 GB VRAM | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| M4 Pro 48GB Apple · 48 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| M4 Max 36GB Apple · 36 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| M5 Max 36GB Apple · 36 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| RTX 5090 NVIDIA · 32 GB VRAM | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| Radeon AI Pro R9700 AMD · 32 GB VRAM | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| Intel Arc Pro B70 Intel · 32 GB VRAM | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Fits estimate No measured run |
| M4 32GB Apple · 32 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Tight estimate No measured run |
| M6 32GB Apple · 32 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Tight estimate No measured run |
| RTX 3090 NVIDIA · 24 GB VRAM | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Over estimate No measured run |
| TITAN RTX NVIDIA · 24 GB VRAM | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Over estimate No measured run |
| Radeon RX 7900 XTX AMD · 24 GB VRAM | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Over estimate No measured run |
| M4 Pro 24GB Apple · 24 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Over estimate No measured run |
| M3 Pro 18GB Apple · 18 GB unified | 8.0 GB Fits estimate No measured run | 13.4 GB Tight estimate No measured run | 24.6 GB Over estimate No measured run |
| RTX 4060 Ti 16GB NVIDIA · 16 GB VRAM | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Over estimate No measured run |
| RTX 5060 Ti 16GB NVIDIA · 16 GB VRAM | 8.0 GB Fits estimate No measured run | 13.4 GB Fits estimate No measured run | 24.6 GB Over estimate No measured run |
| RTX 3060 12GB NVIDIA · 12 GB VRAM | 8.0 GB Fits estimate No measured run | 13.4 GB Over estimate No measured run | 24.6 GB Over estimate No measured run |
| RTX 3080 Ti NVIDIA · 12 GB VRAM | 8.0 GB Fits estimate No measured run | 13.4 GB Over estimate No measured run | 24.6 GB Over estimate No measured run |
| RTX 4070 Ti NVIDIA · 12 GB VRAM | 8.0 GB Fits estimate No measured run | 13.4 GB Over estimate No measured run | 24.6 GB Over estimate No measured run |
Measured runs
| Hardware | Quant | tok/s | Runtime | Source |
|---|---|---|---|---|
| RTX 5090 | llama-bench tg128 (quant per result) | 142 | llama.cpp | Source · checked |
| RTX 4090 | QAT Q4_0 (7.0 GB, Google's file) | 99.7 | llama.cpp b11669 (CUDA) | Source · checked · prefill 5072 |
| RTX 4090 | llama-bench tg128 (quant per result) | 98.1 | llama.cpp | Source · checked |
| RTX 4090 | QAT Q4_0 | 97.7 | llama.cpp b8931 | Source · checked · prefill 1743 |
| RTX 5070 Ti | QAT Q4_0 (7.0 GB, Google's file) | 92.5 | llama.cpp b11669 (CUDA) | Source · checked · prefill 2910 |
| RTX 5070 Ti | llama-bench tg128 (quant per result) | 87.6 | llama.cpp | Source · checked |
| RTX 3060 12GB | llama-bench tg128 (quant per result) | 38.5 | llama.cpp | Source · checked |
| RTX 2060 Super | Q4_K_M | 27.7 | llama.cpp server (CUDA, Docker), forced full offload | Source · checked |
| Ryzen AI Max+ 395 64GB | QAT Q4_0 (7.0 GB, Google's file) | 25.9 | llama.cpp b11669 (Vulkan) | Source · checked · prefill 182 |
| M6 32GB | QAT (google/gemma-4-12b-qat) | 18.2 | LM Studio | Source · checked |
| M4 16GB | Q4_K_M | 13.1 | llama.cpp server (Metal) | Source · checked |