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/dolphin-2.9-llama3-8b-GGUF
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA B300 | 290.47 tok/s | 288GB | ✓ Measured |
| NVIDIA B200 | 274.48 tok/s | 192GB | ✓ Measured |
| NVIDIA H200 | 267.96 tok/s | 141GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 262.47 tok/s | 80GB | ✓ Measured |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 239.29 tok/s | 96GB | ✓ Measured |
| NVIDIA A100 80GB SXM4 | 164.09 tok/s | 80GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 153.34 tok/s | 40GB | Est. |
| NVIDIA L40S | 134.94 tok/s | 48GB | ✓ Measured |
| NVIDIA A10G | 87.98 tok/s | 24GB | ✓ Measured |
| NVIDIA L4 | 50.04 tok/s | 24GB | ✓ Measured |
| NVIDIA T4 | 35.1 tok/s | 16GB | ✓ Measured |
NVIDIA B300 is the fastest card we've measured running dolphin-2.9-llama3-8b, at 290.47 tok/s.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs dolphin-2.9-llama3-8b, at 35.1 tok/s.