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: lmstudio-community/Qwen3-4B-Thinking-2507-GGUF
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA B300 | 339.74 tok/s | 288GB | ✓ Measured |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 334.65 tok/s | 96GB | ✓ Measured |
| NVIDIA H200 | 318.94 tok/s | 141GB | ✓ Measured |
| NVIDIA B200 | 317.03 tok/s | 192GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 310.84 tok/s | 80GB | ✓ Measured |
| NVIDIA L40S | 211.88 tok/s | 48GB | ✓ Measured |
| NVIDIA A100 80GB SXM4 | 199.71 tok/s | 80GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 189.89 tok/s | 40GB | Est. |
| NVIDIA A10G | 129.99 tok/s | 24GB | ✓ Measured |
| NVIDIA L4 | 83.12 tok/s | 24GB | ✓ Measured |
| NVIDIA T4 | 67 tok/s | 16GB | ✓ Measured |
NVIDIA B300 is the fastest card we've measured running Qwen3-4B-Thinking-2507, at 339.74 tok/s.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs Qwen3-4B-Thinking-2507, at 67 tok/s.