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/Llama-2-7B-Chat-GGUF
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA B300 | 315.9 tok/s | 288GB | ✓ Measured |
| NVIDIA B200 | 292 tok/s | 192GB | ✓ Measured |
| NVIDIA H200 | 290.81 tok/s | 141GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 284.62 tok/s | 80GB | ✓ Measured |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 263.84 tok/s | 96GB | ✓ Measured |
| NVIDIA A100 80GB SXM4 | 180.09 tok/s | 80GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 172.53 tok/s | 40GB | Est. |
| NVIDIA L40S | 150.35 tok/s | 48GB | ✓ Measured |
| NVIDIA A10G | 96.69 tok/s | 24GB | ✓ Measured |
| NVIDIA L4 | 56.06 tok/s | 24GB | ✓ Measured |
| NVIDIA T4 | 42.98 tok/s | 16GB | ✓ Measured |
NVIDIA B300 is the fastest card we've measured running Llama-2-7B, at 315.9 tok/s.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs Llama-2-7B, at 42.98 tok/s.