Sorry, but Johnny Oil with a shotgun to my head disagrees with your math. and while I never looked at the numbers myself, I am inclined to agree with him that such a plan would be disturbingly “unprofitable”.
-anyone around western spheres of influence in the vicinity of any sort of lever of power to authorize such changes in infrastructure investment
Given the price of RAM and graphics cards, it is obvious that running LLM is at least somewhat limited by the amount of hardware available. So having that hardware sitting idle, except when there is too much solar power, is obviously not economically viable.
I’m saying this because a lot of things seem obvious at first glance, but are still incorrent. Sometimes reality behaves in unexpected ways and one actually has to check some stuff regularly instead of just assuming it.
Like, which scenario is more profitable? Use 100 chips and run them at all times, even when energy prices are high; or buy 200 chips and use them only when energy is cheap? Which scenario is cheaper? One actually has to do the maths, otherwise one cannot just know these things.
Power and grid infrastructure is a limitation that can exceed hardware availability in some regions. Musk has a datacenter with 20-something methane gas generators running throughout the day to power his mini-me sycophantic AI, Grok.
At the cost of a cultural deficit, solar could provide an environmental benefit there during the day.
Gotta admit, didn’t think about that. Maybe the solution was a few guillotines all along. (This solution has its own problem tho, see the Robespierre gambit)
Peak energy production would be a good time to train the damn llms instead of building natural gas power plant I guess.
Sorry, but Johnny Oil with a shotgun to my head disagrees with your math. and while I never looked at the numbers myself, I am inclined to agree with him that such a plan would be disturbingly “unprofitable”.
-anyone around western spheres of influence in the vicinity of any sort of lever of power to authorize such changes in infrastructure investment
Given the price of RAM and graphics cards, it is obvious that running LLM is at least somewhat limited by the amount of hardware available. So having that hardware sitting idle, except when there is too much solar power, is obviously not economically viable.
citation needed
I’m saying this because a lot of things seem obvious at first glance, but are still incorrent. Sometimes reality behaves in unexpected ways and one actually has to check some stuff regularly instead of just assuming it.
Like, which scenario is more profitable? Use 100 chips and run them at all times, even when energy prices are high; or buy 200 chips and use them only when energy is cheap? Which scenario is cheaper? One actually has to do the maths, otherwise one cannot just know these things.
Power and grid infrastructure is a limitation that can exceed hardware availability in some regions. Musk has a datacenter with 20-something methane gas generators running throughout the day to power his mini-me sycophantic AI, Grok.
At the cost of a cultural deficit, solar could provide an environmental benefit there during the day.
Illegal methane gas generators, I might add.
Then you use taxation to change the viability. Make the non renewable energy so expensive for that usage that they’re better just to shutdown.
Gotta admit, didn’t think about that. Maybe the solution was a few guillotines all along. (This solution has its own problem tho, see the Robespierre gambit)
Yes but that would be woke soy and gay. You dont want to get gay woke soy in your ai. Thats against like the entire point of the thing!