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/HuggingFaceTB_SmolLM3-3B-GGUF
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA B300 | 420.3 tok/s | 288GB | ✓ Measured |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 409.73 tok/s | 96GB | ✓ Measured |
| NVIDIA H200 | 404.27 tok/s | 141GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 399.14 tok/s | 80GB | ✓ Measured |
| NVIDIA B200 | 399.05 tok/s | 192GB | ✓ Measured |
| NVIDIA L40S | 272.96 tok/s | 48GB | ✓ Measured |
| NVIDIA A100 80GB SXM4 | 254.6 tok/s | 80GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 247.66 tok/s | 40GB | Est. |
| NVIDIA A10G | 171.66 tok/s | 24GB | ✓ Measured |
| NVIDIA L4 | 110.77 tok/s | 24GB | ✓ Measured |
| NVIDIA T4 | 86.86 tok/s | 16GB | ✓ Measured |
NVIDIA B300 is the fastest card we've measured running SmolLM3-3B, at 420.3 tok/s.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs SmolLM3-3B, at 86.86 tok/s.