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.
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA L4 | 0.97 x vs solo | 24GB | ✓ Measured |
| NVIDIA L40S | 0.95 x vs solo | 48GB | ✓ Measured |
| NVIDIA A10G | 0.92 x vs solo | 24GB | ✓ Measured |
| NVIDIA T4 | 0.9 x vs solo | 16GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 0.73 x vs solo | 80GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 0.69 x vs solo | 40GB | Est. |
NVIDIA L4 is the fastest card we've measured running Qwen2.5 1.5B + 0.5B draft, at 0.97 x vs solo.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs Qwen2.5 1.5B + 0.5B draft, at 0.9 x vs solo.