Spike: Run miners on selected subnets for the Masa foundation

[Luka-Loncar]

Seems can be profitable to run miners on other bittensor subnets - and this could be convenient for the Masa foundation as well.

For instance, SN19 seems more profitable to run, and cover the costs of launching a service from an external party (for instance, runpod).

This card is about finding:

• A list of subnets that can be profitable to run a miner on to start services (e.g. like SN19 supports AI inferencing)
• If it's possible to use our Amazon free AWE credits to run miners for these subnets, and if not, trying to figure out if it's convenient to spawn services like in runpod and expose miners to the subnets.
• Find a setup/configuration that we can use/re-use to replicate the deployments

Acceptance criteria

• [ ] We have a list of subnets that are convenient to run services on
• [ ] We know if it's convenient to use AWE credits to run miners on the identified subnets
• [ ] We know if it's convenient to use runpod to run AI miners (e.g. for SN19)
• [ ] Have a replicable setup that works for runpod
• [ ] Have a replicable setup that works for AWE

Additional context

Seems we had luck before by having profit using runpod to deploy GPUs and start a miner to SN19 https://masafinance.slack.com/archives/C0742F6KXB8/p1719852760560399

Timebox

1 day

[juanmanso]

Triaging this ticket a question that comes up to me is:

 What's does convenient mean to us?

• Is it covering the costs convenient?
• Is it getting X% over the costs convenient?

More over, we need to define what the costs are to be able to compare. Maybe that's up to the stakeholders to figure after the whole set of tasks is done (and not part of devs acceptance criteria).

[juanmanso]

My suggestion is to tailor the spike's steps to completion:

• Find a setup (no matter the cost) that can be replicable using:
  • Either
  • AWS
  • Runpod
  • Other (if suggested)
  • [OUTPUT] Define what's the minimum cost for us to run said setups

• Run other subnets and figure out if the return is greater than these costs

  • [OUTPUT] Table with cost of running vs % return for stakeholders to decide if an actual implementation is due

  Based on this list (which is similar to the one in the description) I think we can break this spike into 2 smaller spike tickets (even break the first one into 2 -- AWS setup and Runpod setup) that are chained together so progress is contained within them and it's easier to track and estimate IMO

[Luka-Loncar]

@teslashibe should we close this maybe or move it to the ice box maybe?