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: Qwen/Qwen2.5-1.5B
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA B200 | 9186.4 serve tok/s | 192GB | ✓ Measured |
| NVIDIA H200 | 7541.6 serve tok/s | 141GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 6794.8 serve tok/s | 80GB | ✓ Measured |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 5786.7 serve tok/s | 96GB | ✓ Measured |
| NVIDIA L40S | 4686.8 serve tok/s | 48GB | ✓ Measured |
| NVIDIA A100 80GB SXM4 | 4564.7 serve tok/s | 80GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 4111.1 serve tok/s | 40GB | Est. |
| NVIDIA A10G | 2702.5 serve tok/s | 24GB | ✓ Measured |
| NVIDIA L4 | 1827 serve tok/s | 24GB | ✓ Measured |
| NVIDIA T4 | 1337.1 serve tok/s | 16GB | ✓ Measured |
NVIDIA B200 is the fastest card we've measured running Qwen2.5 1.5B served, at 9186.4 serve tok/s.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs Qwen2.5 1.5B served, at 1337.1 serve tok/s.