IP Library › Granted Patent US 12,471,238
Granted Patent B2
US 12,471,238 · App. 17/155,959 · Granted Nov 11, 2025

Static data center power balancing and configuration

Inventors: Benjamin D. Faulkner (San Jose, CA); Mini Rawat (San Jose, CA); Tao Li (San Jose, CA); Divya Ramakrishnan (San Jose, CA); Swanand Santosh Bindoo (San Jose, CA); Sreedhar Narayanaswamy (Sunnyvale, CA)
Assignee: NVIDIA Corporation
H05K7/1498G06F1/329G06F9/4893G06F9/5011G06F9/5044G06F9/5094G06T1/20H05K7/1492G06F2209/501Y02D10/00
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 12,471,238
App. No.
17/155,959
Granted
Nov 11, 2025
Kind
B2
Abstract

A system to select graphics processing units (GPUs) to execute a task is disclosed. In at least one embodiment, GPUs are selected based on one or more task parameters and one or more fused parameters.

Claims (44)

1 . A system, comprising:

one or more processors to select two or more graphics processing units (GPUs) to perform a task based, at least in part, upon one or more task parameters associated with the task and one or more fused power coefficients of the two or more GPUs, wherein an operating frequency of each GPU of the two or more GPUs is to be independently tuned to balance power consumption of the two or more GPUs.

2 . The system of claim 1 , wherein two or more selected GPUs are positioned within a common rack.

3 . The system of claim 1 , wherein the one or more fused power coefficients relate operating frequency of the two or more GPUs to power consumption of the two or more GPUs.

4 . The system of claim 1 , wherein two or more selected GPUs form at least a portion of a baseboard.

5 . The system of claim 1 , wherein two or more selected GPUs are positioned within different racks, the different racks operating using different rack operating parameters.

6 . The system of claim 1 , wherein the one or more task parameters include at least one of a power efficiency, a performance level, or a number of GPUs.

7 . The system of claim 1 , wherein the two or more GPUs are statically selected prior to operational power being provided to the two or more GPUs.

8 . A method, comprising:

causing two or more graphics processing units (GPUs) to perform a task based, at least in part, upon one or more task parameters associated with the task and one or more fused power coefficients of the two or more GPUs, wherein an operating frequency of each GPU of the two or more GPUs is to be independently tuned to balance power consumption of the two or more GPUs.

9 . The method of claim 8 , further comprising:

determining a set of GPUs;

determining at least a portion of the set of GPUs have operating parameters insufficient for the one or more task parameters;

determining at least a second portion of the set of GPUs have operating parameters sufficient for the one or more task parameters; and

selecting a subset of GPUs, including at least part of the second portion, to perform the task.

10 . The method of claim 8 , wherein two or more selected GPUs are positioned in different racks, further comprising:

providing a notification to move at least one of the two or more selected GPUs to a common rack with at least one other selected GPU of the two or more selected GPUs.

11 . The method of claim 8 , further comprising:

associating the two or more selected GPUs with the one or more task parameters; and

storing the two or more selected GPUs as a grouping.

12 . The method of claim 11 , further comprising:

receiving a second task and one or more second task parameters;

determining the one or more second task parameters are within a threshold amount of similarity to the one or more task parameters; and

causing the grouping to perform the second task.

13 . The method of claim 8 , further comprising:

adjusting one or more operating parameters of the two or more GPUs, prior to execution of the task, based at least in part on the one or more task parameters.

14 . The method of claim 8 , wherein the one or more task parameters include at least one of a power efficiency, a performance level, or a number of GPUs.

15 . The method of claim 8 , wherein the one or more fused power coefficients include one or more indications of an amount of power to be consumed by each of the two or more GPUs to perform the task at an operating frequency.

16 . The method of claim 8 , wherein the two or more GPUs are statically selected prior to operational power being provided to the two or more GPUs.

17 . A system, comprising:

a first rack including one or more first graphics processing units (GPU);

a second rack including one or more second GPUs; and

a controller including one or more processors to select one or more first GPUs, one or more second GPUs, or a combination thereof, to perform a task based, at least in part, upon one or more task parameters associated with the task and one or more fused power coefficients of the two or more selected GPUs, wherein an operating frequency of each GPU of the two or more selected GPUs is to be independently tuned to balance power consumption of the two or more selected GPUs.

18 . The system of claim 17 , wherein the one or more processors provide a notification to move one or more first GPUs from the first rack to the second rack, based at least in part on the one or more fused power coefficients.

19 . The system of claim 17 , wherein the one or more task parameters include at least one of a power efficiency, a performance level, or a number of GPUs.

20 . The system of claim 17 , wherein the one or more fused power coefficients include respective models for operation of the two or more selected GPUs across multiple applications.

21 . One or more processors comprising:

circuitry to cause two or more graphics processing units (GPUs) to be selected to perform a task based, at least in part, upon one or more task parameters associated with the task and one or more fused power coefficients of the two or more GPUs, wherein an operating frequency of each GPU of the two or more GPUs is to be independently tuned to balance power consumption of the two or more GPUs.

22 . The one or more processors of claim 21 , wherein one or more selected GPUs are positioned within a common rack.

23 . The one or more processors of claim 21 , wherein the one or more fused power coefficients include respective models for operation of the two or more GPUs across multiple applications.

24 . The one or more processors of claim 21 , wherein two or more selected GPUs form at least a portion of a baseboard.

25 . The one or more processors of claim 21 , wherein two or more selected GPUs are positioned within different racks, the different racks operating using different rack operating parameters.

26 . The one or more processors of claim 21 , wherein the one or more task parameters include at least one of a power efficiency, a performance level, or a number of GPUs.

27 . The one or more processors of claim 21 , wherein the two or more GPUs are statically selected prior to operational power being provided to the two or more GPUs.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jan 28, 2021
From: FAULKNER, BENJAMIN D.; RAWAT, MINI; LI, TAO; RAMAKRISHNAN, DIVYA; BINDOO, SWANAND SANTOSH; NARAYANASWAMY, SREEDHAR
To: NVIDIA CORPORATION
Reel/Frame 055068/0063 →
Continuity (1)
Related Publication 20220240408A1 · Jul 28, 2022
References Cited (31)
US 10114431B2 · Garden · 2018 [cited by applicant]
US 10262390B1 · Sun · 2019 [cited by examiner]
US 10325343B1 · Zhao et al. · 2019 [cited by applicant]
US 10725834B2 · Lee · 2020 [cited by applicant]
US 20060059494A1 · Wexler et al. · 2006 [cited by applicant]
US 20090044027A1 · Piazza · 2009 [cited by applicant]
US 20120081373A1 · Li · 2012 [cited by examiner]
US 20130074077A1 · Miller · 2013 [cited by applicant]
US 20150074435A1 · Jane et al. · 2015 [cited by applicant]
US 20160140259A1 · Ponamgi · 2016 [cited by examiner]
US 20160274636A1 · Kim · 2016 [cited by examiner]
US 20160321776A1 · Zou · 2016 [cited by examiner]
US 20170277246A1 · Nicholson · 2017 [cited by examiner]
US 20190042383A1 · Querbach et al. · 2019 [cited by applicant]
US 20190163540A1 · Lee · 2019 [cited by examiner]
US 20190195924A1 · Gomi · 2019 [cited by examiner]
US 20190361678A1 · Kim · 2019 [cited by examiner]
US 20200084919A1 · Tsai · 2020 [cited by examiner]
US 20200183788A1 · Dreyfuss · 2020 [cited by examiner]
US 20210117202A1 · Levit-Gurevich · 2021 [cited by examiner]
US 20220245487A1 · Shen · 2022 [cited by examiner]
CN 103229123A · 2013 [cited by applicant]
CN 104106053A · 2014 [cited by applicant]
WO 2019153188 · 2019 [cited by applicant]
Combined Search and Examination Report issued in GB Application No. GB2200797.5, dated Aug. 9, 2022. [cited by applicant]
Office Action for Chinese Application No. 202210072801.3, mailed Aug. 14, 2024, 17 pages. [cited by applicant]
Office Action for Chinese Application No. 202210072801.3, mailed Feb. 6, 2024, 14 pages. [cited by applicant]
Office Action for United Kingdom Application No. GB2200797.5, mailed Aug. 9, 2024, 4 pages. [cited by applicant]
Office Action for United Kingdom Application No. GB2200797.5, mailed Dec. 20, 2023, 5 pages. [cited by applicant]
Zhang et al., “Research on GPU General-Purpose Computing Virtualization Technology for Multi-tasking,” Computer Engineering and Science, 35(11): Nov. 2013, 7 pages. [cited by applicant]
Examination Report for United Kingdom Application No. GB2200797.5 mailed Jun. 10, 2025, 3 pages. [cited by applicant]