Hacker News Reader

>We're able to maintain fast rendering while charging ~50% less than most render farms by crowdsourcing. We leverage underutilized graphics cards from hardware owners across the globe to keep rentaflop fast and affordable.

Seti@home meets Blender.

I wonder if that would work for machine learning...

View on HN

Crowdsourcing for machine learning is certainly possible, and it's one of the things we've thought about a lot. One issue that current Deep Learning algorithms, for instance, have with crowdsourcing is that they require periodic communication between devices to sync neural network weights. This is typically not an issue when devices are physically located in the same place with little network latency but becomes a bottleneck when the devices are scattered across the globe. With that being said, there are a few startups I'm aware of that are doing this sort of thing.

There is https://vast.ai which is a generic GPU crowdsourcing platform.

There are limitations that make it difficult to set up their machines as infrastructure however.

Reply on news.ycombinator.com