DETAILED ACTION
Notice of Pre-AIA or AIA Status
1. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
2. Claims 1–7, 9–16, and 18–22 are pending for examination in the reply filed on 07/01/2026.
Claims 8 and 17 are cancelled.
And claims 21–22 are NEW.
Examiner’s Remarks
3. Examiner refers to and explicitly cites particular pages, sections, figures, paragraphs or columns and lines in the references as applied to Applicant’s claims to the extent practicable to streamline prosecution.
Although the cited portions of the references are representative of the best teachings in the art and are applied to meet the specific limitations of the claims, other uncited but related teachings of the references may be equally applicable as well. It is respectfully requested that, in preparing responses to the rejections, the Applicant fully considers not only the cited portions of the references, but also the references in their entirety, as potentially teaching, suggesting or rendering obvious all or one or more aspects of the claimed invention.
Abbreviations
4. Where appropriate, the following abbreviations will be used when referencing Applicant’s submissions and specific teachings of the reference(s):
i. figure / figures: Fig. / Figs.
ii. column / columns: Col. / Cols.
iii. page / pages: p. / pp.
References Cited
5. (A) Gutierrez et al., US 2019/0310864 A1 (“Gutierrez”).
(B) Sarood et al., US 2020/0403855 A1 (“Sarood”).
(C) Acharya et al., US 2022/0044350 A1 (“Acharya”).
Gutierrez, Sarood, and Acharya were cited in the previous Office action.
Notice re prior art available under both pre-AIA and AIA
6. In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102 of this title, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
A.
7. Claims 1–5, 7, 10–14, 16, and 19–20, and 22 are rejected under 35 U.S.C. 103 as being unpatentable over (A) Gutierrez.
See “References Cited” section, above, for full citations of references.
8. Regarding claim 1, (A) Gutierrez teaches/suggests the invention substantially as claimed, including:
“A method for performing computing work, the method comprising:
comparing a workload measure threshold to application-provided information indicating a workload measure for a group of execution instances”
(¶ 55: During this operation, the controller functional block compares the behavior to at least one threshold and determines, based on the comparison, which precision level is to be used;
¶ 56: a behavior such as an amount of time to complete executing the test portion, a value or average value of some or all of the results of the test portion, a number of iterations to reach a specified result value, an amount of electrical power used, etc. can be COMPARED TO A CORRESPONDING THRESHOLD …
When the behavior exceeds the threshold, a first precision level can be selected to be used for executing the remainder of the workload. When the behavior does not exceed the threshold, a second precision level can be selected to be used for executing the remaining portion of the workload. (The first precision level or the second precision level may be the precision level at which the test portion of the workload was executed.)
As described above, in some embodiments, the first precision level is a higher-precision precision level than the second precision level. In these embodiments, the threshold can be set to represent a point at which a benefit of executing the remaining portion of the workload at the first/higher-precision precision level outweighs the costs in terms of electrical power consumed, heat generated, time taken, etc.;
¶ 51: While the computational functional block executes the test portion of the workload at the precision level, the controller functional block monitors a behavior of the computational functional block … behavior can include any hardware or software values that may be used for the decision, such as a time taken for performing operations or operation completion rate, a communication bus bandwidth consumed … an amount of electrical power used, an amount of heat generated, etc.;
¶ 13: electronic device also includes a controller functional block that dynamically configures the computational functional block to use precision levels from a set of precision levels when executing WORKLOADS;
¶ 17: controller functional block receives a hint or indication that a specified precision level is to be used for a given workload and/or for the above-described check of the one or more precision levels;
¶ 21: computational functional block 102 includes circuit elements that execute workloads using one of two or more precision levels, with each precision level being characterized at least in part by a number of bits ( or “bit width”) used for operands and/or results during operations;
¶ 43: the software application workloads that are recognized by the controller functional block may have specified patterns or types of instructions, operations, etc. that are amenable to the use of different precision levels. For example, processing operations for training a neural network to perform classification tasks ( e.g., speech recognition, etc.) and/or using the trained neural network to perform classification tasks may be amenable to use of precision levels lower than those implemented in program code. Generally, this means that hardware entities in the electronic device, i.e., the computational functional block and the controller functional block, can override or otherwise control the behavior of program code as the program code is executed by computational functional block (i.e., at the “hardware” level) so that a precision level different than a precision level specified in the program code is used);
“BASED ON THE COMPARING, identifying a first set of execution instances, of the group of execution instances, to operate at a … (higher) precision and a second set of execution instances, of the group of execution instances, to operate at a reduced (lower) precision,”
(¶ 18: By dynamically selecting precision levels to which the computational functional block is to be configured as described herein, the described embodiments may be able to use lower-precision precision levels in situations where higher precisions ( e.g., statically specified precision levels in software applications) might otherwise be used. This can help to avoid the consumption of electrical power, avoid the generation of heat, and improve the speed at which operations are performed, which can, in turn, improve the overall power consumption and performance of the electronic device. These embodiments are therefore an improvement over existing electronic devices that do not include the capability to dynamically set precision levels for workloads;
¶ 20: Computational functional block 102 is a functional block that performs operations for executing workloads such as performing operations for hardware entities ( e.g., other functional blocks, etc.) or software entities (e.g., software applications, etc.);
¶ 40: when the computational functional block is a GPGPU or a compute unit in a GPGPU and the electronic device includes multiple versions of a kernel, the controller functional block may cause the computational functional block to execute the workload using a version of the kernel that is compiled or otherwise arranged to use operands and/or results of the selected precision level);
“wherein both the first set of execution instances and the second set of execution instances specify operation at the … (higher) precision”
(¶ 47: controller functional block may receive or otherwise acquire, from another hardware or software entity, a request, hint, or instruction to select a particular precision level and may select the precision level based thereon … a software application may provide a hint or other indication, such as the above-described static specification or a dedicated hint, that the controller functional block should use a particular precision level for executing a workload for the software application;
¶ 17: controller functional block receives a hint or indication that a specified precision level is to be used for a given workload and/or for the above-described check of the one or more precision levels;
¶ 13: The precision levels, and thus the bit widths used for operands and/or results, can include any bit width that can be operated on by the computational functional block, from 1 bit to 256 bits and more);
“operating the first set of execution instances at the … (higher) precision and the second set of execution instances at the reduced precision”
(¶ 21: executing workloads using precision levels/bit widths for operands and/or results that include 16 bit values (e.g., floating point values, etc.), 32 bit values, 64 bit values etc. In these embodiments, computational functional block 02 is dynamically configurable to use a specified one of the precision levels for executing a given workload as described herein).
Gutierrez do not explicitly describe the higher precision level as a “NORMAL precision.”
Gutierrez however teaches in the Background that some workloads can “include operations for which a precision of operands and/or results is specified in advance or “statically” specified. For example, a software application executed by a chip (e.g., a microprocessor, a compute unit, etc.) may be compiled with a specification of a precision such as 64 bit floating point operands and results. In some cases, such software applications include operations that do not necessarily require the full specified precision … When performing such classification tasks, neural networks may be able to produce results that are sufficiently accurate without requiring the full specified precision be used” (paragraph 2) and in paragraph 46 that “a precision level, such as a precision level specified ( or “statically” specified) by a programmer for a software application in which the above described
neural network is implemented, may be unnecessarily high. In other words, the programmer that programmed the neural network may have specified a precision level that is higher than necessary to generate correct results from the neural network).
Accordingly, it is inherent in or would have been obvious to a person of ordinary skill in the art in view of Gutierrez’s teachings/disclosure that for at least some workloads the higher precision level (e.g. 64-bit) is reasonably understood as a “NORMAL” or the initial, statically specified precision for workloads requiring a higher or full precision (which may be dynamically, and conditionally (based on certain behaviors/conditions), switched to using a lower or “reduced” precision level).
9. Regarding claim 2, Gutierrez teaches or suggests:
“wherein identifying the second set of execution instances includes identifying one or more execution instances, of the group of execution instances, for which the workload measure greater than the workload measure threshold” (with second set of execution instances operate at a REDUCED precision)
(¶ 56: a behavior such as an amount of time to complete executing the test portion, a value or average value of some or all of the results of the test portion, a number of iterations to reach a specified result value, an amount of electrical power used, etc. can be COMPARED TO A CORRESPONDING THRESHOLD …
When the behavior exceeds the threshold, a first precision level can be selected to be used for executing the remainder of the workload. When the behavior does not exceed the threshold, a second precision level can be selected to be used for executing the remaining portion of the workload. (The first precision level or the second precision level may be the precision level at which the test portion of the workload was executed.)
As described above, in some embodiments, the first precision level is a higher-precision precision level than the second precision level. In these embodiments, the threshold can be set to represent a point at which a benefit of executing the remaining portion of the workload at the first/higher-precision precision level outweighs the costs in terms of electrical power consumed, heat generated, time taken, etc.)
The Examiner notes that “wherein identifying the second set of execution instances includes identifying one or more execution instances, of a set of execution instances, that have a workload measure greater than a threshold” is an obvious and corresponding embodiment of Gutierrez’s teachings for which behaviors such as completion time and electrical power used are compared to a threshold, thus the threshold is set to indicate MORE desirable aspect of 1) using a lower-precision precision level having lower electrical power consumption, heat generation, and faster time to completion, versus 2) using a higher-precision precision level with higher power consumption, higher heat generation, and longer time to completion.
It would have been obvious to incorporate and use this particular embodiment suggested by Gutierrez’s to so as to prioritize faster completion times, lower power consumption (in circumstances where power cost is a major factor), and heat generation (where heat significantly degrades performance or increases probability of failure).
E.g. where power consumption or heat generation exceeds a threshold, switch to using a lower-precision precision level (“second set of execution instances”) to minimize cost and/or performance degradation. But where power consumption or heat generation does NOT exceeds the threshold, keep using the specified higher precision level.
10. Regarding claim 3, Gutierrez teaches or suggests:
“wherein identifying the first set of execution instances includes identifying one or more execution instances, of the group of execution instances, that have a workload measure that is NOT greater than the workload measure threshold” (with first set of execution instances operate at a NORMAL, non-reduced precision)
(¶ 56: a behavior such as an amount of time to complete executing the test portion, a value or average value of some or all of the results of the test portion, a number of iterations to reach a specified result value, an amount of electrical power used, etc. can be COMPARED TO A CORRESPONDING THRESHOLD …
When the behavior exceeds the threshold, a first precision level can be selected to be used for executing the remainder of the workload. When the behavior does not exceed the threshold, a second precision level can be selected to be used for executing the remaining portion of the workload. (The first precision level or the second precision level may be the precision level at which the test portion of the workload was executed.)
As described above, in some embodiments, the first precision level is a higher-precision precision level than the second precision level. In these embodiments, the threshold can be set to represent a point at which a benefit of executing the remaining portion of the workload at the first/higher-precision precision level outweighs the costs in terms of electrical power consumed, heat generated, time taken, etc.)
(The Examiner notes that “wherein identifying the second set of execution instances includes identifying one or more execution instances, of a set of execution instances, that have a workload measure greater than a threshold” is an obvious and corresponding embodiment of Gutierrez’s teachings for which behaviors such as completion time and electrical power used are compared to a threshold, thus the threshold is set to indicate MORE desirable aspect of 1) using a lower-precision precision level having lower electrical power consumption, heat generation, and faster time to completion, versus 2) using a higher-precision precision level with higher power consumption, higher heat generation, and longer time to completion.
It would have been obvious to incorporate and use this particular embodiment suggested by Gutierrez’s to so as to prioritize faster completion times, lower power consumption (in circumstances where power cost is a major factor), and heat generation (where heat significantly degrades performance or increases probability of failure); E.g. where power consumption or heat generation exceeds a threshold, switch to using a lower-precision precision level (“second set of execution instances”) to minimize cost and/or performance degradation.
But where power consumption or heat generation does NOT exceeds the threshold, keep using the specified higher precision level).
11. Regarding claim 4, Gutierrez teaches or suggests:
“wherein the workload measure indicates how busy an execution instance is”
(¶ 56: a behavior such as an amount of time to complete executing the test portion, a value or average value of some or all of the results of the test portion, a number of iterations to reach a specified result value).
12. Regarding claim 5, Gutierrez teaches or suggests:
“wherein the workload measure threshold is for the first set of execution instances and the second set of execution instances”
(¶ 56: a behavior such as an amount of time to complete executing the test portion, a value or average value of some or all of the results of the test portion, a number of iterations to reach a specified result value, an amount of electrical power used, etc. can be compared to a corresponding threshold)
13. Regarding claim 7, Gutierrez teaches or suggests:
“wherein the threshold is an absolute value”
(¶ 56: a behavior such as an amount of time to complete executing the test portion, a value or average value of some or all of the results of the test portion, a number of iterations to reach a specified result value, an AMOUNT of electrical power used, etc. can be compared to a corresponding threshold)
14. Regarding claims 10–14 and 16, they are the corresponding system claims reciting similar limitations of commensurate scope as the method of claims 1–5 and 7, respectively. Therefore, they are rejected on the same basis as claims 1–5 and 7 above, including the following rationale:
Gutierrez teaches or suggests: “a processor configured to: …”
(¶ 64: one or more processors/cores/central processing units (CPUs),) and
“one or more memories storing information for the first set of execution instances and the second set of execution instances”
(¶ 36: switching between precision levels and/or using various precision levels for operands and/or results includes performing operations for ensuring that data (operands, results, etc.) is properly stored in and/or retrieved from memory functional block 106).
15. Regarding claim 22, Gutierrez teaches or suggests:
“wherein execution instances operated at the reduced precision are operated with work-items that utilize fewer bits than execution items operated at the normal precision”
(¶ 14: computational functional block may include a set of circuit elements that are operable at various precision levels via enabling/disabling subsets of circuit elements within the single set of circuit elements, such as an N bit-wide ALU (where N is 256, 128, 64, or another number) that can be configured via disabling respective subsets of circuit elements to operate on operands having numbers of bits less than N;
¶ 48: a precision level corresponds to a bit width or number of bits for operands and/or results of operations;
¶ 26: mechanisms for configuring the bit width of operands and/or results used by the ALU from a highest-precision precision level (e.g., 128 bit) to a lowest-precision precision level (e.g., 16 bit), such as by powering down, halting clocks to, etc. portions of the ALU that are not used in computations for lower precision levels).
16. Regarding claims 19–20, they are the corresponding computer program product claims reciting similar limitations of commensurate scope as the method of claims 1–2, respectively. Therefore, they are rejected on the same basis as claims 1–2 above.
B.
17. Claims 6 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over (A) Gutierrez, as applied to claims 5 and 14 above, and further in view of (B) Sarood.
18. Regarding claim 6, Gutierrez teaches or suggests “the threshold … of the workload measures of the first set of execution instances and the second set of execution instances”
(¶ 56: a behavior such as an amount of time to complete executing the test portion, a value or average value of some or all of the results of the test portion, a number of iterations to reach a specified result value, an amount of electrical power used, etc. can be compared to a corresponding threshold)
Gutierrez do not teach “wherein the threshold is a number of standard deviations above a mean of the workload measures of the first set of execution instances and the second set of execution instances.”
(B) Sarood, in the context of Gutierrez’s teachings, however teaches or suggests
“wherein the threshold is a number of standard deviations above a mean of the workload measures of the first set of execution instances and the second set of execution instances”
(¶ 48: test whether one or more operational parameters are operating within a particular number of variances or standard deviations from their historical averages. Similarly, the second criterion may be defined to determine whether one or more operational parameters are operating within a second particular number of variances or standard deviations from their historical averages. Depending on the operational parameter being monitored, a positive numerical change may indicate either a degradation ( e.g. CPU utilization) or an improvement ( e.g. throughput). Thus, the rules engine may be configured to define the first and second criterion based on the characteristics of each of the operational parameters measured, such that the first criterion is met when an overall degradation is detected;
¶ 57: the first threshold may be set at a first number of variances or standard deviations above a mean or median value for an operational parameter. The second threshold may be set to a second number of variances or standard deviations above or below mean or median value for the operational parameters;
Fig. 3: graph showing historical operational parameter values, e.g., SLE, relative to possible changes in those values resulting).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of (B) Sarood with those of (A) Gutierrez to set the threshold as a number of (performance) standard deviations from the historical or expected average value for an operational parameter (e.g. power consumption, heat generation) . The motivation or advantage to do so is to implement resource monitoring rules for determining performance or operational degradation (where switching to a lower precision level is more beneficial).
19. Regarding claim 15, it is the corresponding system claim reciting similar limitations of commensurate scope as the method of claim 6. Therefore, it is rejected on the same basis as claim 6 above.
C.
20. Claims 9, 18 and 21 are rejected under 35 U.S.C. 103 as being unpatentable over (A) Gutierrez, as applied to claims 2 and 11 above, and further in view of (C) Acharya.
21. Regarding claim 9, Gutierrez do not teach “wherein the workload measure is based on performance counters.”
(C) Acharya, in the context of Gutierrez’s teachings, however teaches or suggests
“wherein the workload measure is based on performance counters”
(¶ 22: performance counter data generated by the performance counters, such as the average temperature of the device, the rate of change (RoC) of the average temperature of the device 100, the peak instantaneous power consumption of the device 100 over a given time period, the average power consumption of the device 100 over a given time period, the RoC of the average power consumption of the device ….).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of (C) Acharya with those of (A) Gutierrez to use performance counters for tracking/monitoring behavior data. The motivation or advantage to do so is to allow for activity tracking and logging at various locations, modules, and/or circuit elements of the computational device.
22. Regarding claim 18, it is the corresponding system claim reciting similar limitations of commensurate scope as the method of claim 9. Therefore, it is rejected on the same basis as claim 9 above.
23. Regarding claim 21, it is the same method claim as the method of claim 9. Therefore, it is rejected on the same basis as claim 9 above.
Response to Arguments
24. Applicant’s arguments with respect to the claims have been considered but are moot because the arguments do not apply to any of the newly applied teachings or references being used in the current rejection.
In the Remarks, the Applicant also contends the following:
a. There is no mention in Gutierrez of the above features of claims 1, 10, and 19, including
comparing a workload measure threshold to application-provided information indicating a workload measure for a group of execution instances.
The Examiner disagrees.
As to (a), as applied in the rejections, Gutierrez teaches in these features in at least
¶ 55: During this operation, the controller functional block compares the behavior to at least one threshold and determines, based on the comparison, which precision level is to be used;
¶ 56: a behavior such as an amount of time to complete executing the test portion, a value or average value of some or all of the results of the test portion, a number of iterations to reach a specified result value, an amount of electrical power used, etc. can be COMPARED TO A CORRESPONDING THRESHOLD …
When the behavior exceeds the threshold, a first precision level can be selected to be used for executing the remainder of the workload. When the behavior does not exceed the threshold, a second precision level can be selected to be used for executing the remaining portion of the workload. (The first precision level or the second precision level may be the precision level at which the test portion of the workload was executed.)
As described above, in some embodiments, the first precision level is a higher-precision precision level than the second precision level. In these embodiments, the threshold can be set to represent a point at which a benefit of executing the remaining portion of the workload at the first/higher-precision precision level outweighs the costs in terms of electrical power consumed, heat generated, time taken, etc.;
¶ 51: While the computational functional block executes the test portion of the workload at the precision level, the controller functional block monitors a behavior of the computational functional block … behavior can include any hardware or software values that may be used for the decision, such as a time taken for performing operations or operation completion rate, a communication bus bandwidth consumed … an amount of electrical power used, an amount of heat generated, etc.;
¶ 13: electronic device also includes a controller functional block that dynamically configures the computational functional block to use precision levels from a set of precision levels when executing WORKLOADS;
¶ 17: controller functional block receives a hint or indication that a specified precision level is to be used for a given workload and/or for the above-described check of the one or more precision levels;
¶ 21: computational functional block 102 includes circuit elements that execute workloads using one of two or more precision levels, with each precision level being characterized at least in part by a number of bits ( or “bit width”) used for operands and/or results during operations;
¶ 43: the software application workloads that are recognized by the controller functional block may have specified patterns or types of instructions, operations, etc. that are amenable to the use of different precision levels. For example, processing operations for training a neural network to perform classification tasks ( e.g., speech recognition, etc.) and/or using the trained neural network to perform classification tasks may be amenable to use of precision levels lower than those implemented in program code ….;
¶ 18: By dynamically selecting precision levels to which the computational functional block is to be configured as described herein, the described embodiments may be able to use lower-precision precision levels in situations where higher precisions ( e.g., statically specified precision levels in software applications) might otherwise be used. This can help to avoid the consumption of electrical power, avoid the generation of heat, and improve the speed at which operations are performed, which can, in turn, improve the overall power consumption and performance of the electronic device. These embodiments are therefore an improvement over existing electronic devices that do not include the capability to dynamically set precision levels for workloads.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
(a) GADELRAB et al., US 2021/0279635 A1, teaching adaptively executing machine learning models on a computing device in a high accuracy mode.
(B) Sadowski et al., US 2018/0113678 A1, teaching dynamic variable precision computation.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BENJAMIN C WU whose telephone number is (571)270-5906. The examiner can normally be reached Monday through Friday, 8:30 A.M. to 5:00 P.M..
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Aimee J. Li can be reached on (571)272-4169. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/BENJAMIN C WU/Primary Examiner, Art Unit 2195
September 3, 2026
Prosecution Insights