IP Library › Granted Patent US 12,645,505
Granted Patent B2
US 12,645,505 · App. 18/102,963 · Granted Jun 2, 2026

Apparatus and method with computing system control based on reinforcement learning for operating in high performance / efficiency mode

Inventors: Hyung-Dal Kwon (Suwon-si, KR); Seongbeom Kim (Suwon-si, KR)
Assignee: SAMSUNG ELECTRONICS CO., LTD.
G06F9/5094G06F1/206G06F1/3206G06F1/324G06F1/329G06F1/3296G06F9/30189G06F9/4837G06F9/505G06F9/5072G06F9/5083G06N3/092G06F9/45558G06F2009/4557G06F9/5027G06F9/5077G06F2209/501Y02D10/00
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,645,505
App. No.
18/102,963
Granted
Jun 2, 2026
Kind
B2
Abstract

An apparatus with computing system control includes: a receiver configured to receive an operation mode for controlling a plurality of components constituting a computing system; and a processor configured to: determine a parameter for controlling the plurality of components based on the operation mode, a time limit for processing an operation of the plurality of components, and an operation time of the plurality of components; distribute a workload for the plurality of components based on the parameter; and process the operation based on the distributed workload.

Claims (88)

1 . An apparatus with computing system control, the apparatus comprising:

a processor configured to:

determine one or more parameters for controlling a plurality of components constituting a computing system based on whether an operation mode for controlling the plurality of components is a high performance mode or a high efficiency mode and on an operation time to complete processing of one or more workloads using the plurality of components, and by performing reinforcement learning on a neural network based on temperatures of the plurality of components, operation voltages of the plurality of components, operation frequencies of the plurality of components, power consumptions of the plurality of components, peak performances of the plurality of components, and power efficiencies of the plurality of components;

in response to the operation mode being the high performance mode, compare the power consumptions of the plurality of components;

in response to the operation mode being the high efficiency mode, compare the operation voltages according to the operation frequencies of the plurality of components;

distribute the one or more workloads to the plurality of components based on the one or more parameters and on a result of the comparing of the power consumptions or a result of the comparing of the operation voltages according to the operation mode; and

process the distributed one or more workloads on the plurality of components.

2 . The apparatus of claim 1 , wherein, for the determining of the one or more parameters, the processor is configured to:

determine the one or more parameters for controlling the plurality of components based on the operation mode being the high performance mode; and

determine the one or more parameters for controlling the plurality of components based on the operation mode being the high efficiency mode.

3 . The apparatus of claim 1 , wherein the plurality of components comprises any one or any combination of any two or more of an operation component, a storage component, a communication component, and a power component.

4 . The apparatus of claim 1 , wherein, for the determining of the one or more parameters, the processor is configured to determine either one or both of an operation voltage and an operation frequency for controlling the plurality of components based on the operation mode and the operation time of the plurality of components.

5 . The apparatus of claim 1 , wherein, for the determining of the one or more parameters, the processor is configured to determine the one or more parameters based on the power consumptions of the plurality of components, performances of the plurality of components, and the temperatures of the plurality of components.

6 . The apparatus of claim 1 , wherein, for the distributing of the one or more workloads, the processor is configured to, in response to the operation mode being the high performance mode:

assign different workloads of the one or more workloads to the plurality of components based on the result of the comparing of the power consumption.

7 . The apparatus of claim 1 , wherein for the determining of the one or more parameters, the processor is configured to perform either one of:

changing an operation mode of a portion of the plurality of components in response to a command to change the operation mode from a user of the computing system in real time; and

automatically changing an operation mode of a portion of the plurality of components based on a predetermined value.

8 . The apparatus of claim 6 , wherein

the plurality of components comprises a first component and a second component,

the different workloads comprise a first workload, and a second workload having an amount of computation greater than an amount of computation of the first workload, and

for the assigning of the different workloads, the processor is configured to, in response to power consumption of the first component being greater than power consumption of the second component:

assign the first workload to the first component; and

assign the second workload to the second component.

9 . The apparatus of claim 1 , wherein, for the distributing of the one or more workloads, the processor is configured to, in response to the operation mode being the high efficiency mode:

assign different workloads of the one or more workloads to the plurality of components based on a result of the comparing of the operation voltages.

10 . The apparatus of claim 9 , wherein

the plurality of components comprises a first component and a second component,

the different workloads comprise a first workload, and a second workload with an amount of computation greater than an amount of computation of the first workload, and

for the assigning of the different workloads, the processor is configured to, in response to an operation voltage of the first component being less than an operation voltage of the second component at a same operation frequency:

assign the second workload to the first component; and

assign the first workload to the second component.

11 . The apparatus of claim 1 , wherein

the distributed workload comprises a first workload, and a second workload with an amount of computation greater than an amount of computation of the first workload, and

for the distributing of the workload, the processor is configured to:

generate a first cluster and a second cluster by clustering the plurality of components based on dynamic voltage and frequency scaling (DVFS) characteristics of the plurality of components;

assign the second workload to a cluster excellent in the DVFS characteristics among the first cluster and the second cluster; and

assign the first workload to the other cluster.

12 . A processor-implemented method with computing system control, the method comprising:

receiving an operation mode for controlling a plurality of components constituting a computing system;

determining one or more parameters for controlling the plurality of components based on whether the operation mode is a high performance mode or a high efficiency mode and on an operation time to complete processing of one or more workloads using the plurality of components, and by performing reinforcement learning on a neural network based on temperatures of the plurality of components, operation voltages of the plurality of components, operation frequencies of the plurality of components, power consumptions of the plurality of components, peak performances of the plurality of components, and power efficiencies of the plurality of components;

in response to the operation mode being the high performance mode, comparing the power consumptions of the plurality of components;

in response to the operation mode being the high efficiency mode, comparing the operation voltages according to the operation frequencies of the plurality of components;

distributing the one or more workloads to the plurality of components based on the one or more parameters and on a result of the comparing of the power consumptions or a result of the comparing of the operation voltages according to the operation mode; and

processing the distributed one or more workloads on the plurality of components.

13 . The method of claim 12 , wherein the determining of the one or more parameters comprises selectively, based on the received operation mode:

determining the one or more parameters for controlling the plurality of components based on the operation mode being the high performance mode; and

determining the one or more parameters for controlling the plurality of components based on the operation mode being the high efficiency mode.

14 . The method of claim 12 , wherein the plurality of components comprises any one or any combination of any two or more of an operation component, a storage component, a communication component, and a power component.

15 . The method of claim 12 , wherein the determining of the one or more parameters comprises determining either one or both of an operation voltage and an operation frequency for controlling the plurality of components based on the operation mode and the operation time of the plurality of components.

16 . The method of claim 12 , wherein the determining of the one or more parameters comprises determining the one or more parameters based on the power consumptions of the plurality of components, performances of the plurality of components, and the temperatures of the plurality of components.

17 . The method of claim 12 , wherein the distributing of the one or more workloads comprises, in response to the operation mode being the high performance mode:

assigning different workloads of the one or more workloads to the plurality of components based on a result of the comparing of the power consumptions.

18 . The method of claim 12 , wherein

the receiving of the operation mode comprises receiving a command to change the operation mode from a user of the computing system in real time, and

the determining of the one or more parameters comprises either one of:

changing an operation mode of a portion of the plurality of components in response to the command; and

automatically changing an operation mode of a portion of the plurality of components based on a predetermined value.

19 . The method of claim 17 , wherein

the plurality of components comprises a first component and a second component,

the different workloads comprise a first workload, and a second workload having an amount of computation greater than an amount of computation of the first workload, and

the assigning of the different workloads to the plurality of components comprises, in response to power consumption of the first component being greater than power consumption of the second component:

assigning the first workload to the first component; and

assigning the second workload to the second component.

20 . The method of claim 12 , wherein the distributing of the one or more workloads comprises, in response to the operation mode being the high efficiency mode:

assigning different workloads of the one or more workloads to the plurality of components based on the result of the comparing of the operation voltages.

21 . The method of claim 20 , wherein

the plurality of components comprises a first component and a second component,

the different workloads comprise a first workload, and a second workload having an amount of computation greater than an amount of computation of the first workload, and

the assigning of the different workloads to the plurality of components comprises, in response to an operation voltage of the first component being less than an operation voltage of the second component at a same operation frequency:

assigning the second workload to the first component; and

assigning the first workload to the second component.

22 . The method of claim 12 , wherein

the distributed workload comprises a first workload, and a second workload with an amount of computation greater than an amount of computation of the first workload, and

the distributing of the workload comprises:

generating a first cluster and a second cluster by clustering the plurality of components based on dynamic voltage and frequency scaling (DVFS) characteristics of the plurality of components;

assigning the second workload to a cluster excellent in the DVFS characteristics among the first cluster and the second cluster; and

assigning the first workload to the other cluster.

23 . A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, configure the processor to perform the method of claim 12 .

24 . A processor-implemented method with computing system control, the method comprising:

determining a first operation parameter of a first component and a second operation parameter of a second component based on whether an operation mode for controlling a computing system comprising the first component and the second component is a high performance mode or a high efficiency mode, and by performing reinforcement learning on a neural network based on temperatures of the first and second components, operation voltages of the first and second components, operation frequencies of the first and second components, power consumptions of the first and second components, peak performances of the first and second components, and power efficiencies of the first and second components;

in response to the operation mode being the high performance mode, comparing the power consumptions of the first and second components;

in response to the operation mode being the high efficiency mode, comparing the operation voltages according to the operation frequencies of the first and second components;

based on a comparison result of the power consumptions or a comparison result of the operation voltages, assigning a first workload to the first component and a second workload, having a greater amount of computation than the first workload, to the second component, in response to a value of the first operation parameter being greater than a value of the second operation parameter according to the operation mode; and

processing the first workload the second workload using the first component and the second component based on the assigned first workload and the assigned second workload on the plurality of components.

25 . The method of claim 24 , wherein the determining of the first operation parameter and the second operation parameter comprises:

determining the first operation parameter and the second operation parameter to be power consumptions in response to the operation mode being the high performance mode; and

determining the first operation parameter and the second operation parameter to be operation voltages in response to the operation mode being the high efficiency mode.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jan 30, 2023
From: KWON, HYUNG-DAL; KIM, SEONGBEOM
To: SAMSUNG ELECTRONICS CO., LTD.
Reel/Frame 062530/0711 →
Priority Claims (1)
KR 10-2022-0101576 · Aug 12, 2022 · national
Continuity (1)
Related Publication 20240054024A1 · Feb 15, 2024
References Cited (28)
US 8473724B1 · Kenville · 2013 [cited by examiner]
US 20080313640A1 · Liu · 2008 [cited by examiner]
US 20090144568A1 · Fung · 2009 [cited by examiner]
US 20110022870A1 · McGrane · 2011 [cited by examiner]
US 20140115363A1 · Tu · 2014 [cited by examiner]
US 20140215241A1 · Yoon · 2014 [cited by examiner]
US 20140281647A1 · Bodas · 2014 [cited by examiner]
US 20160239065A1 · Lee · 2016 [cited by examiner]
US 20170322616A1 · Hsiao et al. · 2017 [cited by applicant]
US 20180024860A1 · Balle · 2018 [cited by examiner]
US 20180095520A1 · Rajwan · 2018 [cited by examiner]
US 20180373287A1 · Al-Rawi · 2018 [cited by examiner]
US 20190265777A1 · Lee · 2019 [cited by examiner]
US 20190370086A1 · Heilper et al. · 2019 [cited by applicant]
US 20200050920A1 · Idgunji et al. · 2020 [cited by applicant]
US 20200301860A1 · Hsu et al. · 2020 [cited by applicant]
US 20210081027A1 · Park et al. · 2021 [cited by applicant]
US 20210081209A1 · Meswani et al. · 2021 [cited by applicant]
US 20220091653A1 · Subramony · 2022 [cited by examiner]
US 20230217253A1 · Palermo · 2023 [cited by examiner]
US 20230305905A1 · Chen · 2023 [cited by examiner]
KR 1020210101337A · 2021 [cited by applicant]
KR 1020230047872A · 2023 [cited by applicant]
WO WO2005098579A2 · 2005 [cited by applicant]
WO WO2014065970A1 · 2014 [cited by applicant]
Extended European search report issued on Jan. 5, 2024, in counterpart European Patent Application No. 23179512.1 (12 pages). [cited by applicant]
Gocht, Andreas, et al. “Q-learning inspired self-tuning for energy efficiency in hpc.” [cited by applicant]
Tang, Zhenheng, et al. “The impact of GPU DVFS on the energy and performance of deep learning: An empirical study.” [cited by applicant]