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/gemma-2-2b-it-GGUF
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA B300 | 417.84 tok/s | 288GB | ✓ Measured |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 417.07 tok/s | 96GB | ✓ Measured |
| NVIDIA H200 | 398.48 tok/s | 141GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 392.26 tok/s | 80GB | ✓ Measured |
| NVIDIA B200 | 391.94 tok/s | 192GB | ✓ Measured |
| NVIDIA L40S | 278.1 tok/s | 48GB | ✓ Measured |
| NVIDIA A100 80GB SXM4 | 244.52 tok/s | 80GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 234.68 tok/s | 40GB | Est. |
| NVIDIA A10G | 175.03 tok/s | 24GB | ✓ Measured |
| NVIDIA L4 | 114.17 tok/s | 24GB | ✓ Measured |
| NVIDIA T4 | 92.13 tok/s | 16GB | ✓ Measured |
NVIDIA B300 is the fastest card we've measured running gemma-2-2b, at 417.84 tok/s.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs gemma-2-2b, at 92.13 tok/s.