Hacker News Reader

$600k is an order of magnitude less than what it cost to train GPT-3 or DALL-E 2.

When that figure came out, the popular talking point was how cheap Stable Diffusion cost to make and how easily a well-funded competitor could create their own custom variant.

View on HN

> how easily a well-funded competitor could create their own custom variant.

Unsurprisingly, Pornhub is already using machine learning. Their first big project is colorizing and upscaling classic porn.[1] (Moderately NSFW). As they point out, they have plenty of training data.

[1] https://www.pornhub.com/art/remastured

$600k is also list price for the GPU time spent. As an investor in a GPU cloud company the actual cost was probably way less than that.

They absolutely didn't pay list price. Even the rookiest AWS account manager is going to be able to get a discount on $600k of compute.

You can get up to $100k in startup credits on AWS just by asking. They certainly cut a deal for Stable Diffusion on top of that.

The trouble isn't renting those GPUs but it was finding that number of GPUs closely connected to each other for the duration of the training process.

The founder said that this is not quite as possible with most public clouds and that it is easier to buy the GPUs.

Reply on news.ycombinator.com