Estimated by Hydra based on public benchmarks, research papers, and cloud provider environmental impact reports.
GPT-4 (OpenAI)
Carbon emissions tons: 1587
Water waste liters: 5200000
Energy consumption kWh: 10500000
Data center location: Iowa, USA (Microsoft Azure)
Renewable energy percentage: 64
Hardware used: NVIDIA A100 GPUs
Model parameters: 175000000000
Training hours: 7200
PaLM 2 (Google DeepMind)
Carbon emissions tons: 1890
Water waste liters: 5700000
Energy consumption kWh: 11800000
Data center location: Oregon, USA (Google Cloud)
Renewable energy percentage: 95
Hardware used: TPU v4 Pods
Model parameters: 540000000000
Training hours: 8000
Claude 3 (Anthropic)
Carbon emissions tons: 1270
Water waste liters: 4300000
Energy consumption kWh: 8600000
Data center location: Virginia, USA (AWS)
Renewable energy percentage: 50
Hardware used: NVIDIA A100 GPUs
Model parameters: 120000000000
Training hours: 6000
LLaMA 2 (Meta)
Carbon emissions tons: 1340
Water waste liters: 4450000
Energy consumption kWh: 9000000
Data center location: Iowa, USA (Meta Cloud)
Renewable energy percentage: 100
Hardware used: Custom AI Accelerators
Model parameters: 65000000000
Training hours: 5000
Bard (Google)
Carbon emissions tons: 1650
Water waste liters: 4900000
Energy consumption kWh: 11500000
Data center location: California, USA (Google Cloud)
Renewable energy percentage: 98
Hardware used: TPU v4 Pods
Model parameters: 600000000000
Training hours: 8500
DALL-E 3 (OpenAI)
Carbon emissions tons: 1120
Water waste liters: 3800000
Energy consumption kWh: 8000000
Data center location: Iowa, USA (Microsoft Azure)
Renewable energy percentage: 64
Hardware used: NVIDIA A100 GPUs
Model parameters: 150000000000
Training hours: 4500
Midjourney
Carbon emissions tons: 1020
Water waste liters: 3600000
Energy consumption kWh: 7600000
Data center location: Virginia, USA (AWS)
Renewable energy percentage: 50
Hardware used: Custom GPUs
Model parameters: Proprietary
Training hours: 4000
Hugging Face Transformers
Carbon emissions tons: 910
Water waste liters: 3100000
Energy consumption kWh: 7100000
Data center location: Frankfurt, Germany (AWS)
Renewable energy percentage: 75
Hardware used: NVIDIA A100 GPUs
Model parameters: 110000000000
Training hours: 5500
Stability AI (Stable Diffusion)
Carbon emissions tons: 970
Water waste liters: 3300000
Energy consumption kWh: 7400000
Data center location: Helsinki, Finland (Local Provider)
Renewable energy percentage: 100
Hardware used: Custom GPUs
Model parameters: 130000000000
Training hours: 4800
Cohere AI
Carbon emissions tons: 870
Water waste liters: 3000000
Energy consumption kWh: 6900000
Data center location: Montreal, Canada (AWS)
Renewable energy percentage: 90
Hardware used: NVIDIA A100 GPUs
Model parameters: 120000000000
Training hours: 6200