this post was submitted on 14 Jul 2024
79 points (88.3% liked)
Fuck AI
1147 readers
560 users here now
"We did it, Patrick! We made a technological breakthrough!"
A place for all those who loathe AI to discuss things, post articles, and ridicule the AI hype. Proud supporter of working people. And proud booer of SXSW 2024.
founded 6 months ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Have you got a source for that?
A quick search suggests that generating an image consumes between 0.01 and 0.29 kWhs (quite a range, so let's hit the middle and use 0.14kWh), while playing Cyberpunk on a PS5 pulls about 200W, so ~0.2kWh per hour.
Seems pretty comparable, assuming you only generate a few images an hour... but if you were generating dozens, it seems like you'd overtake gaming pretty quickly.
Edit: I apologise, I took the Google summary of an article at face value. Clicking through to the linked article here actually says per 1,000 image generations which is far lower. Urgh, though actually the article also says 0.01 to 0.29 kWh. I'm just going to find another article. ๐คฆ If you did have a source with numbers, I'd still be interested in seeing it!
Great question! I actually don't really have an official source, it's from my own experience.
I run ComfyUI locally on my own Laptop and generating an image takes 4 seconds, during which my 3070 Laptop GPU uses 80 Watts (the maximum amount of power it can use). It also fully uses one of the 16 threads of my i7-11800H (TDP of 45W). Let's overestimate a bit and say it uses 100% of the CPU (even though in reality it's only 6.25%), which adds 45 watts resulting in 125 watts (or 83 watts if you account for the fact that it only uses one thread).
That's 125 watts for 4 seconds for one image, or about 0.139 WH (0.000139KWH). That would be 7200 images per KWH. Playing one hour of Cyberpunk on a PS5 would be equivalent to me generating 1440 images on my laptop
"Sources":
Oh I forgot to add a disclaimer kinda. I use a model based on SD1.5, which is best for generating new images at 768x512 pixels. Upscaling it to 1536x1024 using AI would take about 4 times as long and thus use 4 times as much energy, but you would only upscale an image after having generated many images and getting one you are satisfied with, so I don't think it makes that big of a difference. Maybe out of 100 pictures I'd upscale one to share it. Though I mostly do AI image generation because I think the technology is really cool. I love toying around with the node system in ComfyUI.