From Words to Watts: Benchmarking the Energy Costs of Large Language Model Inference · HackerPapers discussions

From Words to Watts: Benchmarking the Energy Costs of Large Language Model Inference — HN submission

Wed, 08 Jul 2026 04:33:20 GMT

From Words to Watts: Benchmarking the Energy Costs of LLM Inference (2023)

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

From Words to Watts: Benchmarking the Energy Costs of Large Language Model Inference — mentioned by Ukv

Tue, 28 Jan 2025 15:46:30 GMT

> what do you think has a bigger impact on someone’s quality of life: 1) operating a home or 2) generating AI images? [...] I would argue most people would weigh the convenience of a house a lot higher. To meaningfully compare directly like this, you'd need to choose two things with similar energy usage/carbon footprint. Sure "the convenience of a house" is greater, but a person's usage of an image-generation model also consumes many orders of magnitude less energy. For a balanced comparison: * A full year of LLM usage (5 joules per token[0] * 200 tokens per query * 10 queries per day * 365…

Paper and discussion history. A mention is not an endorsement.

From Words to Watts: Benchmarking the Energy Costs of Large Language Model Inference — mentioned by lhl

Sat, 18 Jan 2025 07:48:22 GMT

I published this as a comment as well, but it's probably worth nothing that the ChatGPT water/power numbers cited (the one that is most widely cited in these discussions) comes from an April 2023 paper (Li et al, arXiv:2304.03271) that estimates water/power usage based off of GPT-3 (175B dense model) numbers published from OpenAI's original GPT-3 2021 paper. From Section 3.3.2 Inference: > As a representative usage scenario for an LLM, we consider a conversation task, which typically includes a CPU-intensive prompt phase that processes the user’s input (a.k.a., prompt) and a memory-intensive…

Paper and discussion history. A mention is not an endorsement.

From Words to Watts: Benchmarking the Energy Costs of Large Language Model Inference — mentioned by Ukv

Fri, 05 Jul 2024 10:53:07 GMT

> At current trend its going to consume ridiculous amount of electricity for no good reason (and make some more CO2). 5 joules per token[0] * 200 tokens per query * 10 queries per day * 365 days per year = 1.014kWh Which, going by the US's energy mix[1], means a year of someone's LLM usage would be about 0.4kg of CO2 - the same as a single cup of coffee[2]. Worth double-checking, since these are just napkin calculations and I could have made a large mistake. If it is correct (within an order of magnitude or two), that seems to me a relatively small amount of energy. Obviously still expensive…

Paper and discussion history. A mention is not an endorsement.

From Words to Watts: Benchmarking the Energy Costs of Large Language Model Inference — HN submission

Fri, 07 Jun 2024 20:17:48 GMT

Benchmarking the Energy Costs of Large Language Model Inference (2023)

2 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.