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/Codestral-22B-v0.1-GGUF
| GPU | Speed | VRAM | Basis |
|---|---|---|---|
| NVIDIA B300 | 107.98 tok/s | 288GB | ✓ Measured |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 107.67 tok/s | 96GB | ✓ Measured |
| NVIDIA B200 | 89.9 tok/s | 192GB | ✓ Measured |
| NVIDIA H200 | 88.03 tok/s | 141GB | ✓ Measured |
| NVIDIA H100 80GB HBM3 | 87.21 tok/s | 80GB | ✓ Measured |
| NVIDIA L40S | 59.53 tok/s | 48GB | ✓ Measured |
| NVIDIA A100 80GB SXM4 | 49.52 tok/s | 80GB | ✓ Measured |
| NVIDIA A100 40GB SXM4 | 48.37 tok/s | 40GB | Est. |
| NVIDIA A10G | 28.01 tok/s | 24GB | ✓ Measured |
| NVIDIA L4 | 17.96 tok/s | 24GB | ✓ Measured |
| NVIDIA T4 | 11.3 tok/s | 16GB | ✓ Measured |
NVIDIA B300 is the fastest card we've measured running Codestral 22B (Q3_K_M), at 107.98 tok/s.
NVIDIA T4 ($2,299 MSRP) is the cheapest card on our bench that runs Codestral 22B (Q3_K_M), at 11.3 tok/s.