The farm grew. There are more and more AI assistants. It was necessary to run LLM models, RAGs, video analytics, forecasts, digital consultants, marketing scenarios and guest flow simulations.
RESTART deployed AI infrastructure and computing power.
There were GPU servers in the underground warm room. LLM models of the farm were running on them, and the heat from the servers was used to heat the guest houses, the mint greenhouse, the capybara winter pavilion, the tea house and the robot room.
The server room was called “Oven of Ilya Muromets v2.0”, because it warmed, thought and did not require firewood.
Solar panels and wind generators worked nearby. During the day the farm was charged with the sun, at night with the wind, and on cloudy days with discipline. During periods of low load, part of the computing resources was used for legally permitted cryptocurrency mining and technological calculations.
| Indicator | Effect |
|---|---|
| Share of recovered server heat | 72% |
| Heating costs in winter | -38% |
| Share of green generation | 64% |
| Cost of an AI request after optimization | -27% |
| GPU cluster disposal | +41% |
| Payback period of the thermal environment | 3.5 years |
- Here it is, a real Russian stove of the future: they don’t lie on it, models are trained on it.
