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/Hermes-3-Llama-3.2-3B-GGUF
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA B300 | 446.78 tok/s | 288GB | ✓ Measured |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 439.65 tok/s | 96GB | ✓ Measured |
| NVIDIA H200 | 430.79 tok/s | 141GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 423.72 tok/s | 80GB | ✓ Measured |
| NVIDIA B200 | 420.6 tok/s | 192GB | ✓ Measured |
| NVIDIA A100 80GB SXM4 | 269.13 tok/s | 80GB | ✓ Measured |
| NVIDIA L40S | 265.87 tok/s | 48GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 261 tok/s | 40GB | Est. |
| NVIDIA A10G | 169.23 tok/s | 24GB | ✓ Measured |
| NVIDIA L4 | 106.14 tok/s | 24GB | ✓ Measured |
| NVIDIA T4 | 84.99 tok/s | 16GB | ✓ Measured |
NVIDIA B300 is the fastest card we've measured running Hermes-3-Llama-3.2-3B, at 446.78 tok/s.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs Hermes-3-Llama-3.2-3B, at 84.99 tok/s.