Text / Chat

LLaMA 3.1 405B Environmental Impact

HeavyEstimated

Largest open-source model ever released — landmark for open AI

Architecture
Dense Transformer (decoder-only)
Parameters
405B
Context
128,000 tokens
Provider
Meta
4.0 Wh
Energy per query
1.6 g
CO₂ per query
8 mL
Water per query
13x more than
vs Google search

Energy per query

4.0 Wh

13x more than a Google search (0.3 Wh)

CO2 per query

1.6 g

Global Average grid (475 gCO₂/kWh)

Water per query

8 mL

~133 queries to fill 1 litre

Processing location

Self-hosted (varies, requires multi-GPU)

Provider

Meta

Category

Text / Chat

Grid carbon intensity

475 g CO2/kWh (27% renewable)

How does LLaMA 3.1 405B compare?

Ranked #106 of 152 models by energy per query

0 Wh1 Wh2 Wh3 Wh4 WhLLaMA 3.2 1BGemini 1.5 ProGPT-4.1 NanoLLaMA 3.1 405BGoogle search (0.3 Wh)

Detailed Breakdown

Energy Consumption

LLaMA 3.1 405B is the largest open-source model ever released, requiring multi-GPU setups (typically 8× A100/H100) for inference. At ~4 Wh per query, it consumes significantly more than smaller variants but matches or exceeds GPT-4-class models in capabilities while being fully open.

Power Source & Carbon

Due to its size, 405B typically runs on cloud GPU clusters. Infrastructure providers like Together AI, Fireworks, and major clouds offer hosted inference. Meta trained the model on its own infrastructure using approximately 60% renewable energy.

Water Usage

At ~7.5 mL per query, the 405B model has a moderate water footprint driven by its multi-GPU inference requirements.

About LLaMA 3.1 405B

LLaMA 3.1 405B is a 405B-parameter text and chat model from Meta, released July 23, 2024. Largest open-source model ever released — landmark for open AI. At 4.0 Wh per query, it uses 13x the energy of a Google search. It runs on a Dense Transformer (decoder-only) architecture.

These figures are estimates derived from hardware specifications and API benchmarks — Meta has not published official energy data for LLaMA 3.1 405B. Actual consumption may vary significantly depending on batching, quantisation, and infrastructure optimisations that we cannot observe from outside.

LLaMA 3.1 405B in Context

36.5 kWh
per year

Your yearly LLaMA 3.1 405B footprint

At 25 queries per day, your annual LLaMA 3.1 405B usage consumes 36.5 kWh — roughly what a fridge uses in a month. That produces 14.6 kg of CO₂.

Key Insights

Open-source weights — can be self-hosted on infrastructure you control

What does your LLaMA 3.1 405B usage cost the planet?

Use our calculator to estimate your personal environmental footprint based on how often you use LLaMA 3.1 405B.

Calculate My Compute

Frequently Asked Questions

How much energy does LLaMA 3.1 405B use per query?

Each LLaMA 3.1 405B query consumes approximately 4.0 Wh of energy. This is 13x more than a traditional Google search (~0.3 Wh).

What is LLaMA 3.1 405B's carbon footprint?

Based on the carbon intensity of Self-hosted (varies, requires multi-GPU), each query produces approximately 1.6 g of CO2. The grid in this region has a carbon intensity of 475 g CO2/kWh with 27% renewable energy.

How much water does LLaMA 3.1 405B use?

Each query consumes approximately 8 mL of water, primarily used for cooling the data centers that process the request.

How does LLaMA 3.1 405B compare to a Google search?

A LLaMA 3.1 405B query uses 13x more than a Google search in terms of energy. A Google search uses approximately 0.3 Wh, while LLaMA 3.1 405B uses 4.0 Wh.

Technical Details

Architecture

Dense Transformer (decoder-only)

Parameters

405B

Context window

128,000 tokens

Release date

2024-07-23

Open source

Yes

Training data cutoff

2024-06