IP Library Granted Patent US 12,638,971
Granted Patent B2
US 12,638,971 · App. 17/820,356 · Granted May 26, 2026

Bi-modal memory idle hysteresis for optimal add-in card accelerator performance and power

Inventors: Marc Beuchat (Folsom, CA); Eric Samson (Folsom, CA); Josh Mastronarde (Sacramento, CA)
Assignee: Intel Corporation
G06F3/0611G06F3/0625G06F3/0655G06F3/0679
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 12,638,971
App. No.
17/820,356
Granted
May 26, 2026
Kind
B2
Abstract

Methods, systems and apparatuses provide for technology that detects an access to memory, wherein the memory is on a discrete graphics device that includes an accelerator, sets an idle hysteresis value of the memory to a first level if the access to the memory is associated with activity in the accelerator, and sets the idle hysteresis value of the memory to a second level if the access to the memory is not associated with the activity in the accelerator, wherein the second level is greater than the first level.

Claims (41)

1 . A discrete graphics device comprising:

a memory;

an accelerator coupled to the memory; and

a power controller including a set of instructions, which when executed by the power controller, cause the power controller to:

detect an access to the memory,

set an idle hysteresis value of the memory to a first level if the access to the memory is associated with activity in the accelerator, wherein the idle hysteresis value of the memory corresponds to a duration during which the memory remains powered on after the memory has no pending access requests to be processed, and

set the idle hysteresis value of the memory to a second level if the access to the memory is not associated with the activity in the accelerator, wherein the second level is greater than the first level.

2 . The discrete graphics device of claim 1 , wherein the instructions, when executed, further cause the discrete graphics device to control a power state of the memory based on the idle hysteresis value.

3 . The discrete graphics device of claim 1 , wherein the idle hysteresis value is set to the first level if the access to the memory is one or more of adjacent to and before a beginning of the activity in the accelerator, during the activity in the accelerator or adjacent to and after an end of the activity in the accelerator.

4 . The discrete graphics device of claim 1 , wherein the first level is approximately twice a power state transition latency of the memory.

5 . The discrete graphics device of claim 1 , wherein the second level is at least twice a workload frame time corresponding to the activity in the accelerator.

6 . At least one non-transitory computer readable storage medium comprising a set of instructions, which when executed by a computing system, cause the computing system to:

detect an access to memory, wherein the memory is on a discrete graphics device that includes an accelerator;

set an idle hysteresis value of the memory to a first level if the access to the memory is associated with activity in the accelerator, wherein the idle hysteresis value of the memory corresponds to a duration during which the memory remains powered on after the memory has no pending access requests to be processed; and

set the idle hysteresis value of the memory to a second level if the access to the memory is not associated with the activity in the accelerator, wherein the second level is greater than the first level.

7 . The at least one non-transitory computer readable storage medium of claim 6 , wherein the instructions, when executed, further cause the computing system to control a power state of the memory based on the idle hysteresis value.

8 . The at least one non-transitory computer readable storage medium of claim 6 , wherein the idle hysteresis value is set to the first level if the access to the memory is adjacent to and before a beginning of the activity in the accelerator.

9 . The at least one non-transitory computer readable storage medium of claim 6 , wherein the idle hysteresis value is set to the first level if the access to the memory is during the activity in the accelerator.

10 . The at least one non-transitory computer readable storage medium of claim 6 , wherein the idle hysteresis value is set to the first level if the access to the memory is adjacent to and after an end of the activity in the accelerator.

11 . The at least one non-transitory computer readable storage medium of claim 6 , wherein the first level is approximately twice a power state transition latency of the memory.

12 . The at least one non-transitory computer readable storage medium of claim 6 , wherein the second level is at least twice a workload frame time corresponding to the activity in the accelerator.

13 . A semiconductor apparatus comprising:

one or more substrates; and

logic coupled to the one or more substrates, wherein the logic is implemented at least partly in one or more of configurable or fixed-functionality hardware, the logic to:

detect an access to memory, wherein the memory is on a discrete graphics device that includes an accelerator;

set an idle hysteresis value of the memory to a first level if the access to the memory is associated with activity in the accelerator, wherein the idle hysteresis value of the memory corresponds to a duration during which the memory remains powered on after the memory has no pending access requests to be processed; and

set the idle hysteresis value of the memory to a second level if the access to the memory is not associated with the activity in the accelerator, wherein the second level is greater than the first level.

14 . The semiconductor apparatus of claim 13 , wherein the logic is to control a power state of the memory based on the idle hysteresis value.

15 . The semiconductor apparatus of claim 13 , wherein the idle hysteresis value is set to the first level if the access to the memory is adjacent to and before a beginning of the activity in the accelerator.

16 . The semiconductor apparatus of claim 13 , wherein the idle hysteresis value is set to the first level if the access to the memory is during the activity in the accelerator.

17 . The semiconductor apparatus of claim 13 , wherein the idle hysteresis value is set to the first level if the access to the memory is adjacent to and after an end of the activity in the accelerator.

18 . The semiconductor apparatus of claim 13 , wherein the first level is approximately twice a power state transition latency of the memory.

19 . The semiconductor apparatus of claim 13 , wherein the second level is at least twice a workload frame time corresponding to the activity in the accelerator.

20 . A method comprising:

detecting an access to memory, wherein the memory is on a discrete graphics device that includes an accelerator;

setting an idle hysteresis value of the memory to a first level if the access to the memory is associated with activity in the accelerator, wherein the idle hysteresis value of the memory corresponds to a duration during which the memory remains powered on after the memory has no pending access requests to be processed; and

setting the idle hysteresis value of the memory to a second level if the access to the memory is not associated with the activity in the accelerator, wherein the second level is greater than the first level.

21 . The method of claim 20 , further including controlling a power state of the memory based on the idle hysteresis value.

22 . The method of claim 20 , wherein the idle hysteresis value is set to the first level if the access to the memory is one or more of adjacent to and before a beginning of the activity in the accelerator, during the activity in the accelerator or adjacent to and after an end of the activity in the accelerator.

23 . The method of claim 20 , wherein the first level is approximately twice a power state transition latency of the memory.

24 . The method of claim 20 , wherein the second level is at least twice a workload frame time corresponding to the activity in the accelerator.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Aug 29, 2022
From: BEUCHAT, MARC; SAMSON, ERIC; MASTRONARDE, JASON
To: INTEL CORPORATION
Reel/Frame 060928/0926 →
Continuity (1)
Related Publication 20240061582A1 · Feb 22, 2024
References Cited (92)
US 4616344A · Noguchi · 1986 [cited by examiner]
US 4942470A · Nishitani · 1990 [cited by examiner]
US 4985838A · Hashiguchi · 1991 [cited by examiner]
US 6098026A · Lee · 2000 [cited by examiner]
US 6119207A · Chee · 2000 [cited by examiner]
US 6657634B1 · Sinclair · 2003 [cited by examiner]
US 8112648B2 · Branover · 2012 [cited by examiner]
US 8607075B2 · Gough · 2013 [cited by examiner]
US 8762923B2 · Hoberman · 2014 [cited by examiner]
US 9015501B2 · Gee · 2015 [cited by examiner]
US 9173224B2 · Cai · 2015 [cited by examiner]
US 9997222B1 · Ku · 2018 [cited by examiner]
US 10178619B1 · Samson · 2019 [cited by examiner]
US 10365841B2 · Elhamias · 2019 [cited by examiner]
US 10754564B2 · Choi · 2020 [cited by examiner]
US 11545968B1 · Chae · 2023 [cited by examiner]
US 12277440B2 · Lee · 2025 [cited by examiner]
US 20010052003A1 · Seki · 2001 [cited by examiner]
US 20030061383A1 · Zilka · 2003 [cited by examiner]
US 20040082362A1 · Peng · 2004 [cited by examiner]
US 20040215371A1 · Samson · 2004 [cited by examiner]
US 20060020835A1 · Samson · 2006 [cited by examiner]
US 20080007561A1 · Thompson · 2008 [cited by examiner]
US 20080023239A1 · Tanaka · 2008 [cited by examiner]
US 20080080468A1 · Shen · 2008 [cited by examiner]
US 20080084271A1 · Jaeger · 2008 [cited by examiner]
US 20080100636A1 · Lai · 2008 [cited by examiner]
US 20080307240A1 · Dahan · 2008 [cited by examiner]
US 20090100275A1 · Chang · 2009 [cited by examiner]
US 20100107166A1 · Topaloglu · 2010 [cited by examiner]
US 20110035498A1 · Shah · 2011 [cited by examiner]
US 20110137495A1 · Sacle · 2011 [cited by examiner]
US 20110296213A1 · Ferlitsch · 2011 [cited by examiner]
US 20120042119A1 · Suzuki · 2012 [cited by examiner]
US 20120069370A1 · Eguchi · 2012 [cited by examiner]
US 20120198465A1 · Hande · 2012 [cited by examiner]
US 20130083611A1 · Ware · 2013 [cited by examiner]
US 20130324151A1 · Lee · 2013 [cited by examiner]
US 20140008346A1 · Zhu · 2014 [cited by examiner]
US 20140129757A1 · Lo · 2014 [cited by examiner]
US 20140171086A1 · Nakamori · 2014 [cited by examiner]
US 20140173298A1 · Bircher · 2014 [cited by examiner]
US 20140176586A1 · Gruber · 2014 [cited by examiner]
US 20140189398A1 · Gorbatov · 2014 [cited by examiner]
US 20140189400A1 · Tsao · 2014 [cited by examiner]
US 20140304566A1 · Henderson · 2014 [cited by examiner]
US 20140328113A1 · Chai · 2014 [cited by examiner]
US 20150208347A1 · Iiyama · 2015 [cited by examiner]
US 20150268713A1 · Jain · 2015 [cited by examiner]
US 20160224272A1 · Kim · 2016 [cited by examiner]
US 20160358303A1 · Juliano · 2016 [cited by examiner]
US 20170031754A1 · Chinnakkonda Vidyapoornachary · 2017 [cited by examiner]
US 20170057040A1 · Rzasa · 2017 [cited by examiner]
US 20170228196A1 · Alavoine · 2017 [cited by examiner]
US 20180129269A1 · Garg · 2018 [cited by examiner]
US 20180129270A1 · Garg · 2018 [cited by examiner]
US 20180242104A1 · Lee · 2018 [cited by examiner]
US 20190041941A1 · Chamarty · 2019 [cited by examiner]
US 20190042316A1 · Luo · 2019 [cited by examiner]
US 20190050048A1 · Kang · 2019 [cited by examiner]
US 20190097764A1 · Nishiguchi · 2019 [cited by examiner]
US 20190163255A1 · Dewey · 2019 [cited by examiner]
US 20190170079A1 · Sasaki · 2019 [cited by examiner]
US 20190213776A1 · Banerjee · 2019 [cited by examiner]
US 20190215769A1 · Samson · 2019 [cited by examiner]
US 20190235615A1 · Shows · 2019 [cited by examiner]
US 20190286214A1 · Pabalkar · 2019 [cited by examiner]
US 20190359050A1 · Komuro · 2019 [cited by examiner]
US 20190369707A1 · Iwamoto · 2019 [cited by examiner]
US 20200117265A1 · Balakrishnan · 2020 [cited by examiner]
US 20200212943A1 · Banin · 2020 [cited by examiner]
US 20210055774A1 · Rodgers · 2021 [cited by examiner]
US 20210074086A1 · Korenaga · 2021 [cited by examiner]
US 20210081313A1 · Jang · 2021 [cited by examiner]
US 20210096921A1 · Banerjee · 2021 [cited by examiner]
US 20210193198A1 · Johnson · 2021 [cited by examiner]
US 20210200297A1 · Gada · 2021 [cited by examiner]
US 20220004436A1 · Di Febbo · 2022 [cited by examiner]
US 20220071535A1 · Jernigan · 2022 [cited by examiner]
US 20220171551A1 · Adsure · 2022 [cited by examiner]
US 20220188016A1 · Dai · 2022 [cited by examiner]
US 20220216708A1 · Wee · 2022 [cited by examiner]
US 20220229566A1 · Benisty · 2022 [cited by examiner]
US 20230090567A1 · Tsien · 2023 [cited by examiner]
US 20230101640A1 · Doctor · 2023 [cited by examiner]
US 20230140640A1 · Gurtovoy · 2023 [cited by examiner]
US 20230195644A1 · Tsien · 2023 [cited by examiner]
US 20230236653A1 · Kang · 2023 [cited by examiner]
US 20230281129A1 · Iizawa · 2023 [cited by examiner]
US 20230333889A1 · Kim · 2023 [cited by examiner]
US 20230359265A1 · Kumar · 2023 [cited by examiner]
US 20240184354A1 · Lin · 2024 [cited by examiner]