Google Gemma-3-12B-IT VRAM Requirements & GPU Benchmarks
Google Gemma-3-12B-IT (12.2B total, 12.2B active, 48L, 8 KV heads, vocab 262144) requires 10.75 GB in 4-bit Q4_K_M, 15.26 GB in FP8, and 26.21 GB in FP16 at 8K context.
COMPARADOR CLOUD GPU Y CALCULADORA ALQUILAR VS COMPRAR
¿No tienes suficiente VRAM? Alquila una GPU en la nube en 60 segundos (Google Gemma-3-12B-IT: 10.75 GB in 4-bit)
Recommended Cloud Tier: RTX 4090 (24GB) · Full FP16 Needs: 26.21 GB
Meses para amortizar compra vs alquiler: ~32 months (vs $1,699 hardware)
| Cloud Provider | RTX 4090 (24GB) | A100 (80GB) | H100 (80GB) | Est. Monthly Cost (4h/day) | Free Credit / Trial | Deploy |
|---|---|---|---|---|---|---|
| RunPod | $0.44 / hr | $1.39 / hr | $2.39 / hr | $52.80 / mo | Instant PyTorch / vLLM Templates + Serverless GPU | Desplegar en RunPod → |
| Vast.ai | $0.29 / hr | $0.95 / hr | $1.79 / hr | $34.80 / mo | Save 50–70% on RTX 4090 / 5090 / A100 rentals | Alquilar GPU Spot → |
| Vultr Cloud GPU | — | $1.79 / hr | $2.69 / hr | $214.80 / mo (A100) | $300 Free Credit | Reclamar $300 Gratis → |
| DigitalOcean GPU | — | — | $2.99 / hr | $358.80 / mo (H100) | $200 Free Credit (60d) | Obtener $200 Gratis → |
| GPU Hardware | VRAM | Bandwidth | FP16 (26.21GB) | Q4_K_M (10.75GB) | Decode / Prefill Speed | Report |
|---|---|---|---|---|---|---|
| NVIDIA RTX 5090 | 32 GB | 1792 GB/s | Sí | Sí (34%) | ~167.2 / 4465 tok/s | Calcular VRAM y Velocidad Exacta → |
| NVIDIA RTX 4090 | 24 GB | 1008 GB/s | No | Sí (45%) | ~94 / 3521 tok/s | Calcular VRAM y Velocidad Exacta → |
| NVIDIA RTX 5080 | 16 GB | 960 GB/s | No | Sí (67%) | ~89.6 / 2402 tok/s | Calcular VRAM y Velocidad Exacta → |
| NVIDIA RTX 4080 Super | 16 GB | 736 GB/s | No | Sí (67%) | ~68.7 / 2225 tok/s | Calcular VRAM y Velocidad Exacta → |
| NVIDIA RTX 4070 Ti Super | 16 GB | 672 GB/s | No | Sí (67%) | ~62.7 / 1880 tok/s | Calcular VRAM y Velocidad Exacta → |
| NVIDIA RTX 3090 | 24 GB | 936 GB/s | No | Sí (45%) | ~87.3 / 1513 tok/s | Calcular VRAM y Velocidad Exacta → |
| NVIDIA RTX 4060 Ti 16GB | 16 GB | 288 GB/s | No | Sí (67%) | ~26.9 / 940 tok/s | Calcular VRAM y Velocidad Exacta → |
| NVIDIA RTX 3060 12GB | 12 GB | 360 GB/s | No | Sí (90%) | ~33.6 / 541 tok/s | Calcular VRAM y Velocidad Exacta → |
| NVIDIA RTX 6000 Ada | 48 GB | 960 GB/s | Sí | Sí (22%) | ~89.6 / 3883 tok/s | Calcular VRAM y Velocidad Exacta → |
| NVIDIA RTX A6000 | 48 GB | 768 GB/s | Sí | Sí (22%) | ~71.7 / 1650 tok/s | Calcular VRAM y Velocidad Exacta → |
| NVIDIA L40S 48GB | 48 GB | 864 GB/s | Sí | Sí (22%) | ~80.6 / 7715 tok/s | Calcular VRAM y Velocidad Exacta → |
| NVIDIA A100 80GB SXM | 80 GB | 2039 GB/s | Sí | Sí (13%) | ~190.2 / 6649 tok/s | Calcular VRAM y Velocidad Exacta → |
| NVIDIA H100 80GB SXM | 80 GB | 3350 GB/s | Sí | Sí (13%) | ~312.5 / 18000 tok/s | Calcular VRAM y Velocidad Exacta → |
| NVIDIA H200 141GB SXM | 141 GB | 4800 GB/s | Sí | Sí (8%) | ~350 / 18000 tok/s | Calcular VRAM y Velocidad Exacta → |
| NVIDIA Blackwell B200 192GB | 192 GB | 8000 GB/s | Sí | Sí (6%) | ~350 / 18000 tok/s | Calcular VRAM y Velocidad Exacta → |
| AMD Instinct MI300X 192GB | 192 GB | 5300 GB/s | Sí | Sí (6%) | ~350 / 18000 tok/s | Calcular VRAM y Velocidad Exacta → |
| 2x NVIDIA RTX 3090 (48GB) | 48 GB | 936 GB/s | Sí | Sí (22%) | ~87.3 / 3026 tok/s | Calcular VRAM y Velocidad Exacta → |
| 2x NVIDIA RTX 4090 (48GB) | 48 GB | 1008 GB/s | Sí | Sí (22%) | ~94 / 7041 tok/s | Calcular VRAM y Velocidad Exacta → |
| 4x NVIDIA A100 80GB (320GB) | 320 GB | 2039 GB/s | Sí | Sí (3%) | ~190.2 / 18000 tok/s | Calcular VRAM y Velocidad Exacta → |
| 8x NVIDIA H100 80GB (640GB) | 640 GB | 3350 GB/s | Sí | Sí (2%) | ~312.5 / 18000 tok/s | Calcular VRAM y Velocidad Exacta → |
| Apple M4 Max (128GB Unified) | 128 GB | 546 GB/s | Sí | Sí (8%) | ~55.7 / 784 tok/s | Calcular VRAM y Velocidad Exacta → |
| Apple M3 Ultra Mac Studio (512GB) | 512 GB | 819 GB/s | Sí | Sí (2%) | ~83.6 / 1155 tok/s | Calcular VRAM y Velocidad Exacta → |
| Apple M4 Pro Mac Mini (64GB) | 64 GB | 273 GB/s | Sí | Sí (17%) | ~27.9 / 392 tok/s | Calcular VRAM y Velocidad Exacta → |