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: bartowski/cognitivecomputations_Dolphin-Mistral-24B-Venice-Edition-GGUF
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA B300 | 121.87 tok/s | 288GB | ✓ Measured |
| NVIDIA B200 | 113.8 tok/s | 192GB | ✓ Measured |
| NVIDIA H200 | 109.2 tok/s | 141GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 105.63 tok/s | 80GB | ✓ Measured |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 93.71 tok/s | 96GB | ✓ Measured |
| NVIDIA A100 80GB SXM4 | 65.77 tok/s | 80GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 62.56 tok/s | 40GB | Est. |
| NVIDIA L40S | 48.43 tok/s | 48GB | ✓ Measured |
| NVIDIA A10G | 31 tok/s | 24GB | ✓ Measured |
| NVIDIA L4 | 17.2 tok/s | 24GB | ✓ Measured |
| NVIDIA T4 | 11.33 tok/s | 16GB | ✓ Measured |
NVIDIA B300 is the fastest card we've measured running Dolphin-Mistral-24B-Venice-Edition, at 121.87 tok/s.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs Dolphin-Mistral-24B-Venice-Edition, at 11.33 tok/s.