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: unsloth/gemma-3-270m-it-GGUF
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA H200 | 967.35 tok/s | 141GB | ✓ Measured |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 965 tok/s | 96GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 961.29 tok/s | 80GB | ✓ Measured |
| NVIDIA B300 | 926.67 tok/s | 288GB | ✓ Measured |
| NVIDIA B200 | 889.52 tok/s | 192GB | ✓ Measured |
| NVIDIA L40S | 822.52 tok/s | 48GB | ✓ Measured |
| NVIDIA A100 80GB SXM4 | 617.58 tok/s | 80GB | ✓ Measured |
| NVIDIA A10G | 597.49 tok/s | 24GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 574.24 tok/s | 40GB | Est. |
| NVIDIA L4 | 493.31 tok/s | 24GB | ✓ Measured |
| NVIDIA T4 | 362.8 tok/s | 16GB | ✓ Measured |
NVIDIA H200 is the fastest card we've measured running gemma-3-270m, at 967.35 tok/s.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs gemma-3-270m, at 362.8 tok/s.