Abstract
From Alexandria, Virginia, NewsRx journalists report that a patent by the inventors Bolotin, Evgeny (Kenmore, WA, US), Dalal, Gal (Haifa, IL), Fu, Yaosheng (Santa Clara, CA, US), Mannor, Shie (Haifa, IL), Nellans, David (Round Rock, TX, US), Yan, Zi (Somerville, MA, US), filed on March 31, 2022, was published online on January 23, 2024. The patent's assignee for patent number 11880261 is NVIDIA Corporation (Santa Clara, California, United States). News editors obtained the following quote from the background information supplied by the inventors: “Computing platforms, or systems, are increasingly judged not just by their performance at a specific power envelope, but by overall energy efficiency or energy per task. The overall carbon footprint of these systems for processing workloads is becoming increasingly important. As such, power delivery and thermal conditions associated with processing workloads are scrutinized. “The thermal conditions in computing systems fluctuate over relatively long periods (seconds or even minutes) and the thermal state at any given moment is dictated by the underlying workload(s). As a result, the sequence of power management choices taken over many relatively short power management phases (each typically controlled at the granularity of milliseconds) contributes to the long term thermal and power conditions of the system. Making the proper power management choices for the various phases, however, can be a complex process that involves continuous system-level decision-making. Because of the inherent problem complexity, current processor power management mechanisms for compute applications often leverage relatively simple heuristics to boost the clock frequency when hardware utilization is high and decrease the clock frequency when hardware utilization is low.”
NETL
NSTL