Hacker News Reader

Comment by freediver | Hacker News Reader

Built WooRender to allow easy access to GPU compute. Basically leasing my deep learning server with 10x 1080ti cards for $100/day.

https://woorender.com

View on HN

Pretty cool! How about turning this into a service where other people can lease their own GPU hardware? Either through your website or by offering it as a sort of turnkey solution.

Already exists: https://vast.ai/

Yes if it already exists lets certainly not make it

I've though about this and the unit economics look good, especially with so many GPUs sitting idle now due to crypto situation. There have been some attempts to build this (vectordash.com and vast.ai are examples) but not sure they achieved type of scale I think it is possible. There are about 5 million GPUs bought to mine crypto that are mainly doing nothing today. Getting 1% of them to join a marketplace would make for GPU compute that probably exceeds AWS, GCE and Azure combined. Happy to chat more about this if anyone is interested, my contact is in profile.

The picture of all ten of those cards together is breathtaking.

With the upcoming consumer RTX launch in a couple days, I really hope you've gotten your money's worth out of them. It's at least been a really long product cycle since their release.

How the heck do you fit 10 cards into one server??? Super cool btw.

Read more about the server here: https://www.servethehome.com/deeplearning11-10x-nvidia-gtx-1...

Wow this is very cool. Can you share how much you're making?

Launched two weeks ago and had 5 days rented out so far. There is demand for this but a lot of capital expenditure. The server itself is $20k+ and then you have hosting. I am using a DC in bay area and it is $1200/month for server. So I can't say this is a business, more like an experiment right now.

Sounds really cool. Best of luck! :)

This site takes 45 seconds to load.

It took less than 2s for me on an iPhone 6S. (Your comment made me click the link because I really wanted to know what was taking 45s)

Took less than five seconds to load for me, your connection must be horrible.

Reply on news.ycombinator.com