Making an image with generative AI uses as much energy as charging your phone

technologyreview.com

4 points·annabyrd··6 comments

The actual paper [0] says approx 4860 joules per image, so 70 images is the same as boiling a litre of water.

Charging a phone takes 20k joules.

[0] https://arxiv.org/pdf/2311.16863.pdf

> The actual paper [0] says approx 4860 joules per image, so 70 images is the same as boiling a litre of water.

And even that's high, judging by how fast I can generate images with normal settings on my local PC with SDXL and my computers power supply capacity.

> We ran all of our experiments on a single NVIDIA A100-SXM4-80GB GPU hosted on Amazon Web Services and used the Code Carbon package to measure both the energy consumed and the carbon emitted during inference.

> [on] (AWS’s us-west-2), which is based in Oregon and has an average carbon intensity of 297.6 grams of 2eq per kWh3.

The A100 can consume 700W

> The A100 can consume 700W

No, it can consume 400W [0], and it should be pumping images, in either case, much faster than my 3080Ti laptop if you manage to fully utilize it (and if you don’t, then its not going to consume that much power, either.)

EDIT: corrected link to correct version of datasheet for the A100-SXM4-80Gb referenced in the paper.

[0] https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Cent...

ack, I was thinking of the H100

The AI industry was hoping you wouldn't notice the exorbitant energy requirements until after this hype cycle had peaked.

Reply on news.ycombinator.com