IP Library Granted Patent US 12,461,585
Granted Patent B2
US 12,461,585 · App. 18/185,008 · Granted Nov 4, 2025

Granular GPU DVFS with execution unit partial powerdown

Inventors: Kenneth Daxer (Sunnyvale, CA); Stephen H. Gunther (Beaverton, OR); Michael N. Derr (El Dorado Hills, CA); Eric Samson (Folsom, CA)
Assignee: Intel Corporation
G06F1/3287G06F1/329
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 12,461,585
App. No.
18/185,008
Granted
Nov 4, 2025
Kind
B2
Abstract

Described herein, in one embodiment, are techniques to facilitate the partial powerdown of sub-components of an execution unit or other graphics processing resource based on the workload to be executed. In another embodiment, granular dynamic voltage and frequency scaling is enabled in which the voltage and frequency of groups of processing resources within a graphics processor can be separately scaled.

Claims (38)

1 . A graphics processor comprising:

a processing resource including a plurality of components, the plurality of components including fixed function resource components and execution resource components, each execution resource component including a vector engine, a matrix engine, and multiple voltage subdomains; and

circuitry coupled with the plurality of components, the circuitry configured to:

receive a workload for execution;

determine, to execute the workload, a first frequency and a first voltage for the matrix engine of execution resource components of the plurality of components;

determine, to execute the workload, a second frequency and a second voltage for the vector engine of the execution resource components of the plurality of components, wherein the first frequency is less than the second frequency and the first voltage is less than the second voltage; and

configure a first voltage subdomain of an execution resource component with the first frequency and the first voltage and a second voltage subdomain of the execution resource component with the second frequency and the second voltage, the first voltage subdomain including the matrix engine and the second voltage subdomain including the vector engine.

2 . The graphics processor of claim 1 , the circuitry configured to reduce a voltage associated with the first voltage subdomain during execution of the workload.

3 . The graphics processor of claim 2 , the circuitry to reduce the voltage and frequency of the first voltage subdomain to zero in response to a determination that the workload does not include instructions associated with the matrix engine.

4 . The graphics processor of claim 2 , the circuitry to adjust a frequency of the second voltage subdomain during execution of the workload independently of the frequency of the first voltage subdomain.

5 . The graphics processor of claim 4 , the circuitry to increase the voltage of the second voltage subdomain during execution of the workload independently of the voltage of the first voltage subdomain.

6 . The graphics processor of claim 5 , wherein the first voltage subdomain is configured to a low power state to facilitate a resumable standby state.

7 . The graphics processor of claim 5 , wherein the processing resource additionally includes a ray tracing core associated with a third voltage subdomain.

8 . The graphics processor of claim 5 , wherein the processing resource additionally includes a video codec associated with a fourth voltage subdomain.

9 . The graphics processor of claim 1 , the circuitry to determine components of the processing resource to be used to execute the workload based on a command buffer associated with the workload and configure a voltage and frequency of the components based on operations to be performed by the workload.

10 . The graphics processor of claim 1 , the circuitry to determine components of the processing resource to be used to execute the workload based on a context type associated with the workload and configure a voltage and frequency of the components based on operations to be performed by the workload.

11 . A method comprising:

receiving a workload for execution at a processing resource of a graphics processor, the processing resource including a plurality of components, the plurality of components including fixed function resource components and execution resource components, each execution resource component including a vector engine, a matrix engine, and multiple voltage subdomains;

determining a first frequency and a first voltage for a matrix engine of execution resource components to execute the workload;

determining a second frequency and a second voltage for a vector engine of the execution resource components to execute the workload, wherein the first frequency is less than the second frequency and the first voltage is less than the second voltage;

configuring a first voltage subdomain of an execution resource component with the first frequency and the first voltage, the first voltage subdomain including the matrix engine; and

configuring a second voltage subdomain of the execution resource component with the second frequency and the second voltage, the second voltage subdomain including the vector engine.

12 . The method of claim 11 , comprising reducing a voltage associated with the first voltage subdomain during execution of the workload.

13 . The method of claim 12 , comprising reducing the voltage and frequency of the first voltage subdomain to zero in response to determining that the workload does not include instructions associated with the matrix engine.

14 . The method of claim 12 , comprising adjusting a frequency of the second voltage subdomain during execution of the workload independently of the frequency of the first voltage subdomain.

15 . The method of claim 14 , comprising increasing the voltage of the second voltage subdomain during execution of the workload independently of the voltage of the first voltage subdomain and configuring the first voltage subdomain to a low power state to facilitate a resumable standby state.

16 . A graphics processing system comprising:

a memory device;

a processing resource coupled with the memory device, the processing resource including a plurality of components, the plurality of components including fixed function resource components and execution resource components, each execution resource component including a vector engine, a matrix engine, and multiple voltage subdomains; and

circuitry coupled with the plurality of components, the circuitry configured to:

receive a workload for execution;

determine, to execute the workload, a first frequency and a first voltage for the matrix engine of execution resource components of the plurality of components;

determine, to execute the workload, a second frequency and a second voltage for the vector engine of the execution resource components of the plurality of components, wherein the first frequency is less than the second frequency and the first voltage is less than the second voltage; and

configure a first voltage subdomain of an execution resource component with the first frequency and the first voltage and a second voltage subdomain of the execution resource component with the second frequency and the second voltage, the first voltage subdomain including the matrix engine and the second voltage subdomain including the vector engine.

17 . The graphics processing system of claim 16 , the circuitry configured to reduce a voltage associated with the first voltage subdomain during execution of the workload.

18 . The graphics processing system of claim 17 , the circuitry to reduce the voltage and frequency of the first voltage subdomain to zero in response to a determination that the workload does not include instructions associated with the matrix engine.

19 . The graphics processing system of claim 16 , the circuitry to determine components of the processing resource to be used to execute the workload based on a command buffer associated with the workload and configure a voltage and frequency of the components based on operations to be performed by the workload.

20 . The graphics processing system of claim 16 , the circuitry to determine components of the processing resource to be used to execute the workload based on a context type associated with the workload and configure a voltage and frequency of the components based on operations to be performed by the workload.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded May 25, 2023
From: DAXER, KENNETH; GUNTHER, STEPHEN H.; DERR, MICHAEL N.; SAMSON, ERIC
To: INTEL CORPORATION
Reel/Frame 063767/0460 →
Continuity (3)
Provisional Application 63321726 · Mar 20, 2022
Provisional Application 63321725 · Mar 20, 2022
Related Publication 20230297159A1 · Sep 21, 2023
References Cited (13)
US 9916636B2 · Li · 2018 [cited by applicant]
US 10186011B2 · Nurvitadhi · 2019 [cited by applicant]
US 10319060B2 · Rao · 2019 [cited by examiner]
US 10359834B2 · Maiyuran et al. · 2019 [cited by applicant]
US 10937125B2 · Kelly · 2021 [cited by examiner]
US 20170293994A1 · Li · 2017 [cited by examiner]
US 20170322617A1 · Maiyuran · 2017 [cited by examiner]
US 20180315158A1 · Nurvitadhi · 2018 [cited by examiner]
US 20190228560A1 · Wald · 2019 [cited by examiner]
US 20220116045A1 · Kumashikar · 2022 [cited by examiner]
US 20220291955A1 · Mrozek · 2022 [cited by applicant]
US 20230334613A1 · Daxer · 2023 [cited by examiner]
Office action for U.S. Appl. No. 18/340,139 mailed Mar. 21, 2025, 8 pages. [cited by applicant]