Monetary Donation Cloud Compute Concept

[Tequila]

I have been thinking about a project idea that might provide another F@H donation path outside of hosting your own hardware and some of the drawbacks (direct power cost, heat, tech knowledge, etc...) for folks who want to donate to various scientific causes F@H tries to solve. Its pretty simple, instead of doing things yourself you donate to a pool of money which 100% goes into cloud processing costs which should reduce some entry pain points (hardware, heat, power cost, sustainability (Google Cloud is carbon neutral), tech knowledge, etc...). Downsides include Cloud Compute cost, but those may be able to be reduced in sales discussions, as well as this basically just being a marketing program to raise funds for more compute resources. This also seems to conflict some with direct donations to the Washington University School of Medicine, but may also provide more engaging with follow up reports on the projects metrics & articles published by the F@H team.

Curious if this idea has been tried and thoughts from the community who currently runs F@H in the cloud and has done very well marketing F@H from a user base level.

Some background on me: I have been folding since the early 2000s off and / on under various usernames/teams but never posted on the forums before.

[foldy]

One option for cheaper cloud compute is to use private rent PCs like at https://vast.ai/console/create/
Because you don't really need a variable cloud data center but individual GPUs to run folding like at home.
You would only need one expert who does setup of machines in cluod and register the donors name and team.
Then the donor pays for the machine and gets his PPD points for his name on FAH website.
I did this setup several times but only for myself. For a fast RTX 2080ti GPU you pay $5 per day and get 3.5 Million PPD.

[Tequila]

Great point. The one thing I like about the cloud is the sustainability of their energy (or at least offset) vs the unknown sources via vast.ai.

That said the 2080ti's at vast.ai seem to run around .209 cents per WU. The closest you can get on Google Cloud is .32 centers per WU on a Tesla T4 (their most economical from the cards I have tested (Still need stats for K80 & P4)). For more comparison the Tesla V100 on Google Cloud cost .956 cents per work unit. I imagine the new A100 is also very expensive per WU favoring speed over cost.

[Ichbin3]

I imagine the new A100 is also very expensive per WU favoring speed over cost.

Some weeks ago I was able to rent 8x A100 as a bunch @vast.ai. Guess how much? I got it for $1.25/h
Every single one was able to make 9Mill PPD per day --> could have made more than 70 Mill/day.
Did spend like $50 in two weeks and got a great push in my stats.
Now there is another bunch of 4x A100, but not so cheap ...