As someone who has tried (hard) to build something like this (seems I'm not the only one), I'll chime in.
This is a great execution of a bad idea. It's the worst kind of bad idea because it seems like a great idea until you try it. It's novel, interesting, timely, and attractive from a technical perspective, but unless your market is tech porn, you're selling to nobody.
This use case is served by countless hosting services, saturated at all levels from cheap VPS's to Amazon clusters to private farms. The market is really efficient here and doesn't afford much overhead. Distributing computation over volunteer users is enormous overhead both for you and your users -- in terms of molding problems to fit the computation model, maintaining the infrastructure, keeping it reliable and performant -- the list is endless. This overhead means you can't compete with the existing players. Forget about being significantly better, which you would need to be to get significant numbers of people to switch.
Quite simply, the math for this doesn't work. Not yet, maybe not ever.
Maybe there's some amazing secret sauce here, but I don't see it. So I offer a warning that a lot of people have tried this idea and failed because they were oblivious to reality.
To add to the market saturation, botnet operators are very interested in monetizing their network now that Bitcoin is getting harder to mine. I never followed through, but I had no problem convincing an open-minded hacker to let me use idle GPUs on infected clients for scientific computation. I didn't even offer to pay.
This is a great execution of a bad idea. It's the worst kind of bad idea because it seems like a great idea until you try it. It's novel, interesting, timely, and attractive from a technical perspective, but unless your market is tech porn, you're selling to nobody.
This use case is served by countless hosting services, saturated at all levels from cheap VPS's to Amazon clusters to private farms. The market is really efficient here and doesn't afford much overhead. Distributing computation over volunteer users is enormous overhead both for you and your users -- in terms of molding problems to fit the computation model, maintaining the infrastructure, keeping it reliable and performant -- the list is endless. This overhead means you can't compete with the existing players. Forget about being significantly better, which you would need to be to get significant numbers of people to switch.
Quite simply, the math for this doesn't work. Not yet, maybe not ever.
Maybe there's some amazing secret sauce here, but I don't see it. So I offer a warning that a lot of people have tried this idea and failed because they were oblivious to reality.