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: TheBloke/Mistral-7B-Instruct-v0.2-GGUF
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA B300 | 306.79 tok/s | 288GB | ✓ Measured |
| NVIDIA B200 | 286.37 tok/s | 192GB | ✓ Measured |
| NVIDIA H200 | 282.32 tok/s | 141GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 276.73 tok/s | 80GB | ✓ Measured |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 253.34 tok/s | 96GB | ✓ Measured |
| NVIDIA A100 80GB SXM4 | 173.8 tok/s | 80GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 166.8 tok/s | 40GB | Est. |
| NVIDIA L40S | 144.82 tok/s | 48GB | ✓ Measured |
| NVIDIA A10G | 92.74 tok/s | 24GB | ✓ Measured |
| NVIDIA L4 | 53.31 tok/s | 24GB | ✓ Measured |
| NVIDIA T4 | 41.31 tok/s | 16GB | ✓ Measured |
NVIDIA B300 is the fastest card we've measured running Mistral-7B-Instruct-v0.2, at 306.79 tok/s.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs Mistral-7B-Instruct-v0.2, at 41.31 tok/s.