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-Coder-0.5B-Instruct-GGUF
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA H200 | 920.36 tok/s | 141GB | ✓ Measured |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 910.13 tok/s | 96GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 892.81 tok/s | 80GB | ✓ Measured |
| NVIDIA B300 | 843.44 tok/s | 288GB | ✓ Measured |
| NVIDIA B200 | 828.48 tok/s | 192GB | ✓ Measured |
| NVIDIA L40S | 715.81 tok/s | 48GB | ✓ Measured |
| NVIDIA A100 80GB SXM4 | 562.59 tok/s | 80GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 542.36 tok/s | 40GB | Est. |
| NVIDIA A10G | 513.36 tok/s | 24GB | ✓ Measured |
| NVIDIA L4 | 397.1 tok/s | 24GB | ✓ Measured |
| NVIDIA T4 | 294.96 tok/s | 16GB | ✓ Measured |
NVIDIA H200 is the fastest card we've measured running Qwen2.5-Coder-0.5B, at 920.36 tok/s.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs Qwen2.5-Coder-0.5B, at 294.96 tok/s.