A model that doesn't fit doesn't run slowly, it doesn't run. Here's every card from our bench, sorted by measured speed.
Weights: bartowski/google_gemma-3-12b-it-GGUF
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 153.93 tok/s | 96GB | ✓ Measured |
| NVIDIA B300 | 149.61 tok/s | 288GB | ✓ Measured |
| NVIDIA H200 | 127.04 tok/s | 141GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 125.92 tok/s | 80GB | ✓ Measured |
| NVIDIA B200 | 125.31 tok/s | 192GB | ✓ Measured |
| NVIDIA L40S | 93.05 tok/s | 48GB | ✓ Measured |
| NVIDIA A100 80GB SXM4 | 71.88 tok/s | 80GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 69.7 tok/s | 40GB | Est. |
| NVIDIA A10G | 46.29 tok/s | 24GB | ✓ Measured |
| NVIDIA L4 | 32.29 tok/s | 24GB | ✓ Measured |
| NVIDIA T4 | 19.79 tok/s | 16GB | ✓ Measured |
NVIDIA RTX PRO 6000 Blackwell Workstation Edition is the fastest card we've measured running Gemma 3 12B (Q3_K_M), at 153.93 tok/s.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs Gemma 3 12B (Q3_K_M), at 19.79 tok/s.