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/ai21labs_AI21-Jamba-Reasoning-3B-GGUF
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 400.02 tok/s | 96GB | ✓ Measured |
| NVIDIA B300 | 380.88 tok/s | 288GB | ✓ Measured |
| NVIDIA H200 | 370.66 tok/s | 141GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 363.09 tok/s | 80GB | ✓ Measured |
| NVIDIA B200 | 360.54 tok/s | 192GB | ✓ Measured |
| NVIDIA L40S | 258.41 tok/s | 48GB | ✓ Measured |
| NVIDIA A100 80GB SXM4 | 239.23 tok/s | 80GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 231.15 tok/s | 40GB | Est. |
| NVIDIA A10G | 166.29 tok/s | 24GB | ✓ Measured |
| NVIDIA L4 | 107.59 tok/s | 24GB | ✓ Measured |
| NVIDIA T4 | 86.91 tok/s | 16GB | ✓ Measured |
NVIDIA RTX PRO 6000 Blackwell Workstation Edition is the fastest card we've measured running AI21-Jamba-Reasoning-3B, at 400.02 tok/s.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs AI21-Jamba-Reasoning-3B, at 86.91 tok/s.