Efficiency Scoreboard
Tokens/kWh
1.2M
Inferences/kWh
8.5K
Cost/Inference
$0.0012
Carbon Emissions
120 kgCO2e
Carbon Mapper
Alerts
High Emissions
Model 'GPT-4-Turbo' in US-East-1 exceeds emission threshold.
Underutilized GPU
GPU usage for 'LLaMA-2-7B' is at 30% capacity.
Cost Spike
Inference costs for 'Claude 3' have increased by 40%.
Recommendations
Optimize Batch Size
Increase batch size for 'LLaMA-2-7B' to improve GPU utilization.
-15 kgCO2e
-$250/month
Shift Workload
Move 'GPT-4-Turbo' workload from US-East-1 to EU-West-3 (lower grid intensity).
-40 kgCO2e
-$50/month
Model Quantization
Apply 8-bit quantization to 'Claude 3' to reduce inference cost.
-5 kgCO2e
-$800/month