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/Qwen2-1.5B-Instruct-GGUF
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 648.49 tok/s | 96GB | ✓ Measured |
| NVIDIA B300 | 583.79 tok/s | 288GB | ✓ Measured |
| NVIDIA H200 | 546.47 tok/s | 141GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 536.52 tok/s | 80GB | ✓ Measured |
| NVIDIA B200 | 454.55 tok/s | 192GB | ✓ Measured |
| NVIDIA L40S | 416.57 tok/s | 48GB | ✓ Measured |
| NVIDIA A100 80GB SXM4 | 333.79 tok/s | 80GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 322.09 tok/s | 40GB | Est. |
| NVIDIA A10G | 265.45 tok/s | 24GB | ✓ Measured |
| NVIDIA L4 | 189.18 tok/s | 24GB | ✓ Measured |
| NVIDIA T4 | 147.68 tok/s | 16GB | ✓ Measured |
NVIDIA RTX PRO 6000 Blackwell Workstation Edition is the fastest card we've measured running Qwen2-1.5B, at 648.49 tok/s.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs Qwen2-1.5B, at 147.68 tok/s.