US 12,093,750 B1
GPU-on-demand system powered by self contained energy source
Karl Andersen, New York, NY (US); and Vitaly Leokumovich, New York, NY (US)
Assigned to Greenlight AI LLC, New York, NY (US)
Filed by Greenlight AI LLC, New York, NY (US)
Filed on Feb. 22, 2024, as Appl. No. 18/584,265.
Int. Cl. G06F 1/00 (2006.01); G05B 13/04 (2006.01); G06F 1/3206 (2019.01); G06F 9/50 (2006.01); G06F 21/60 (2013.01); G06F 1/3203 (2019.01)
CPC G06F 9/5094 (2013.01) [G05B 13/048 (2013.01); G06F 1/3206 (2013.01); G06F 9/5038 (2013.01); G06F 21/606 (2013.01); G06F 1/3203 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A GPU-on-demand system, comprising:
a) a computing device including a graphics processing unit (GPU) and a memory;
b) an Energy Management System (EMS) coupled to the computing device, the EMS configured for regulating power usage and optimizing energy efficiency;
c) a plurality of distributed power resources connected to the EMS and the computing device, the plurality of distributed power resources configured for providing a stable and efficient energy supply;
d) one or more databases configured to store energy metrics gathered from the GPU-on-demand system, the energy metrics including at least energy expenditure of the GPU-on-demand system measured in units of energy;
e) a Large Language Model (LLM) executing on the computing device, the LLM configured for processing the energy metrics stored in the one or more databases so as to generate an energy management plan that is configured to: 1) manage when each of the plurality of distributed power resources shall be used by the GPU-on-demand system, 2) dynamically adjust computing tasks of the computing device based on current energy availability and predicted workload, and 3) maximize GPU utilization while adhering to energy efficiency goals;
f) an API gateway comprising an API coupled to a network connection providing access to a communications network, the API gateway configured for providing external systems secure, on-demand access to the GPU, and wherein the API gateway is configured to manage load on the computing device; and
g) a software module executing on the computing device, the software module configured for managing the GPU-on-demand system according to the energy management plan generated by the LLM by: 1) controlling usage of the plurality of distributed power resources, 2) rate limiting the API gateway and 3) dynamically adjusting the computing tasks of the computing device.