CN118819842A - An artificial intelligence accelerated computing chip - Google Patents

Abstract

The present invention relates to the technical field of computing chips, and discloses an artificial intelligence accelerated computing chip, including: a demand acquisition module for acquiring preset demand conditions, and establishing a retrieval analysis formula according to the preset demand conditions. A crawler module is used to retrieve demand data according to the retrieval analysis formula. A data acquisition module is used to collect demand data, and establish a data set according to the demand data. The processing module is provided with a plurality of processing units, and the processing module is used to perform task data processing on the demand data. The scoring module is used to obtain the data processing capabilities of the plurality of processing units, and to score the processing volume of the processing units according to the data processing capabilities. The control module is used to allocate the amount of data in the data set according to the processing volume scores of the plurality of processing units. The present invention realizes efficient computing and processing of artificial intelligence big data through the coordinated cooperation of various modules, and greatly shortens the processing time of artificial intelligence big data in the existing chip framework.

Description

An artificial intelligence accelerated computing chip

Technical Field

The present invention relates to the technical field of computing chips, and in particular to an artificial intelligence accelerated computing chip.

Background Art

With the rapid development of artificial intelligence technology, the demand for various application scenarios is also growing. However, due to the limitations of traditional architecture, current computing hardware faces the challenge of being unable to perform large-scale multi-task parallel processing when running calculations. This bottleneck not only affects the performance of artificial intelligence applications, but also seriously restricts its practical application in various fields.

At present, the traditional computing hardware architecture mainly focuses on improving the performance of a single core, while ignoring the collaboration between multiple cores. Under this architecture, although the performance of a single processor is very strong, its overall performance is often unsatisfactory when faced with large-scale multi-task parallel processing. However, the shortcomings of this structural design in the face of artificial intelligence applications are obvious. Take machine learning as an example. Machine learning algorithms require a large number of matrix operations and iterative calculations, which are often highly parallel. If the hardware cannot provide sufficient parallel processing capabilities, the execution time of the algorithm will be greatly extended, thereby affecting the efficiency of the entire application.

Therefore, there is an urgent need to invent a hardware device that uses artificial intelligence to accelerate computing, in order to solve the problem that the traditional computing hardware architecture lacks parallel processing capabilities, resulting in long execution time and low computing efficiency during data computing and processing.

Summary of the invention

In view of this, the present invention proposes an artificial intelligence acceleration computing chip, which aims to solve the problem that the traditional computing hardware architecture lacks parallel processing capabilities, resulting in long execution time and low computing efficiency during data computing and processing.

The present invention proposes an artificial intelligence acceleration computing chip, comprising:

The demand acquisition module is used to obtain preset demand conditions and establish a search analysis formula based on the preset demand conditions;

A crawler module, electrically connected to the demand acquisition module, and configured to retrieve demand data according to the retrieval analysis formula;

A data acquisition module, electrically connected to the crawler module, and configured to acquire the required data and establish a data set according to the required data;

A processing module, provided with a plurality of processing units, wherein the processing module is used to perform task data processing on the demand data;

A scoring module, used to obtain the data processing capabilities of the processing units and score the processing volume of the processing units according to the data processing capabilities;

A control module is electrically connected to the processing module and the scoring module respectively, and the control module is used to allocate the amount of data in the data set according to the processing amount scores of the several processing units; wherein,

The control module is also used to obtain the total processing volume scores of the plurality of processing units, and obtain the average processing volume scores of the plurality of processing units based on a formula, the formula being as follows:

q = Q/Y;

Wherein, q is the average score of the processing capacity of the processing units, Q is the total score of the processing capacity of the processing units, and Y is the number of the processing units;

The control module is further configured to allocate the amount of data in the data set according to the relationship between the average score of the processing volume and the processing volume scores of the plurality of processing units.

Furthermore, the data collection module is used to collect the demand data and establish a data set according to the demand data, including:

The data collection module is also used to remove duplicate data in the demand data;

The data collection module is also used to remove abnormal data in the required data for removing duplicate data, wherein the abnormal data includes: undisclosed decrypted data and incompletely disclosed data;

The data collection module is also used to establish a data set using the demand data after removing the duplicate data and abnormal data.

Furthermore, when the scoring module scores the processing volume of the processing unit according to the data processing capability, it includes:

The scoring module is further used to obtain the real-time task processing volume L of the processing unit, and the scoring module is further used to determine whether the real-time task processing volume of the processing unit is in a load state according to the relationship between the real-time task processing volume L and a preset load processing volume L0;

When L≥L0, the control module determines that the real-time task processing amount of the processing unit is in a load state;

When L＜L0, the control module determines that the real-time task processing amount of the processing unit is not in a load state;

The scoring module is further used to obtain processing units that are not in a load state among the processing units, and to score the processing capacity of the processing units that are not in a load state according to the relationship between the real-time task processing capacity L and the preset load processing capacity L0.

Further, the scoring module is also used to obtain a processing unit that is not in a load state among the plurality of processing units, and to score the processing volume of the processing unit that is not in a load state according to the relationship between the real-time task processing volume L and the preset load processing volume L0, including:

The scoring module is further used to obtain a processing volume difference △L between the real-time task processing volume L and a preset load processing volume L0, △L=L-L0, and the scoring module is further used to compare the processing volume difference △L with the preset processing volume difference, and select a relative scoring coefficient according to the comparison result to score the processing volume of the processing unit that is not in the load state;

The scoring module is further used to preset a first preset processing volume difference △L1 and a second preset processing volume difference △L2, and the scoring module is further used to preset a first preset scoring coefficient M1, a second preset scoring coefficient M2 and a third preset scoring coefficient M3, and △L1＜△L2, M1＜M2＜M3;

When △L≤△L1, the first preset scoring coefficient M1 is selected to score the processing capacity of the processing unit that is not in a load state;

When △L1＜△L≤△L2, the second preset scoring coefficient M2 is selected to score the processing capacity of the processing unit that is not in a load state;

When △L＞△L2, the third preset scoring coefficient M3 is selected to score the processing capacity of the processing unit that is not in the load state;

When the scoring module selects the i-th preset scoring coefficient Mi to score the processing capacity of the processing unit that is not in the load state, I=1,2,3, and determines that the processing capacity score of the processing unit that is not in the load state is W, and sets W1=Mi.

Further, when the scoring module selects the i-th preset scoring coefficient Mi to score the processing unit that is not in the load state, and determines that the processing volume score of the processing unit that is not in the load state is W, it includes:

The scoring module is further used to obtain the real-time data processing speed K of the processing unit that is not in a load state, and determine whether the real-time data processing speed of the processing unit that is not in a load state meets the standard data processing speed according to the relationship between the real-time data processing speed K and a preset standard data processing speed K0;

When K=K0, the scoring module determines that the real-time data processing speed of the processing unit meets the standard data processing speed, and does not adjust the processing volume score W of the processing unit that is not in the load state;

When K≠K0, the scoring module determines that the real-time data processing speed of the processing unit does not meet the standard data processing speed, and adjusts the processing volume score W of the processing unit that is not in a load state according to the relationship between the real-time data processing speed K and the preset standard data processing speed K0.

Further, the scoring module determines that the real-time data processing speed of the processing unit does not meet the standard data processing speed, and adjusts the processing volume score W of the processing unit that is not in the load state according to the relationship between the real-time data processing speed K and the preset standard data processing speed K0, including:

The scoring module is further used to obtain a data processing speed difference ΔK between the real-time data processing speed K and a preset standard data processing speed K0, ΔK=K-K0, and the scoring module is further used to compare the processing speed difference ΔK with a preset processing speed difference, and select a corresponding adjustment coefficient according to the comparison result to adjust the processing capacity score W of the processing unit that is not in a load state:

The scoring module is further used to preset a first preset processing speed difference △K1 and a second preset processing speed difference △K2; the scoring module is further used to preset a first preset adjustment coefficient N1, a second preset adjustment coefficient N2 and a third preset adjustment coefficient N3, and △K1＜△K2, N1＜0＜N2＜N3＜0.8;

When △K≤△K1, the third preset adjustment coefficient N3 is selected to adjust the processing capacity score W of the processing unit that is not in a load state;

When ΔK1＜ΔK≤0, the processing capacity score W of the processing unit that is not in the load state is not adjusted;

When 0＜△K≤△K2, the second preset adjustment coefficient N2 is selected to adjust the processing capacity score W of the processing unit that is not in a load state;

When ΔK>ΔK2, the first preset adjustment coefficient N1 is selected to adjust the processing capacity score W of the processing unit that is not in a load state;

When the scoring module selects the i-th preset adjustment coefficient Ni to adjust the processing volume score W of the processing unit that is not in the load state, i=1,2,3, and determines that the adjusted processing volume score of the processing unit that is not in the load state is W1, and sets W1=W*Ni.

Further, when the scoring module selects the i-th preset adjustment coefficient Ni to adjust the processing capacity score W of the processing unit that is not in the load state, and determines that the adjusted processing capacity score of the processing unit that is not in the load state is W1, it includes:

The scoring module is further used to obtain the real-time read/write speed J of the processing unit that is not in a load state, and determine whether to adjust the adjusted processing volume score W1 of the processing unit that is not in a load state according to the relationship between the real-time read/write speed J and the preset standard read/write speed J0;

When J=J0, the scoring module determines not to adjust the processing capacity score W1 of the processing unit that is not in the load state after adjustment;

When J≠J0, the scoring module determines that the real-time read and write speed of the processing unit does not meet the standard read and write speed, and corrects the adjusted processing volume score W1 of the processing unit that is not in a load state based on the relationship between the real-time read and write speed J and the preset standard read and write speed J0.

Further, the scoring module determines that the real-time read/write speed of the processing unit does not meet the standard read/write speed, and corrects the adjusted processing volume score W1 of the processing unit that is not in the load state according to the relationship between the real-time read/write speed J and the preset standard read/write speed J0, including:

The scoring module is further used to obtain a reading/writing speed difference △J between the real-time reading/writing speed J and a preset standard reading/writing speed J0, △J=J-J0, and the scoring module is further used to compare the reading/writing speed difference △J with the preset reading/writing speed difference, and select a corresponding correction coefficient according to the comparison result to correct the adjusted processing volume score W1 of the processing unit that is not in a load state:

The scoring module is further used to preset a first preset reading/writing speed difference △J1 and a second preset reading/writing speed difference △J2; the scoring module is further used to preset a first preset correction coefficient B1, a second preset correction coefficient B2 and a third preset correction coefficient B3, and △J1＜△J2, B1＜0＜B2＜B3＜0.5;

When △J≤△J1, the third preset correction coefficient B3 is selected to correct the adjusted processing capacity score W1 of the processing unit that is not in the load state:

When ΔJ1＜ΔJ≤0, the adjusted processing capacity score W1 of the processing unit that is not in the load state is not corrected;

When 0＜△J≤△J2, the second preset correction coefficient B2 is selected to correct the adjusted processing capacity score W1 of the processing unit that is not in the load state;

When △J＞△J2, the first preset correction coefficient B1 is selected to correct the adjusted processing capacity score W1 of the processing unit that is not in the load state;

When the scoring module selects the i-th preset correction coefficient Bi to correct the adjusted processing volume score W1 of the processing unit that is not in the load state, i=1,2,3, and determines that the corrected processing volume score of the processing unit that is not in the load state is W2, and sets W2=W1*N i.

Furthermore, the control module is further configured to allocate the amount of data in the data set according to the relationship between the average score of the processing volume and the processing volume scores of the plurality of processing units, including:

The control module is further used to obtain the processing volume scores W2 of the plurality of the processing units that are not in a load state, and allocate the data volume in the data set according to the relationship between the processing volume score average score q and the processing volume scores W2 of the plurality of the processing units that are not in a load state;

When W2≥q, the control module allocates the amount of data in the data set to the processing unit according to the relationship between the average processing volume score q and the processing volume score W2 of the processing unit that is not in a load state;

When W2＜q, the control module does not allocate the amount of data in the data set to the processing unit.

Further, when the control module allocates the amount of data in the data set to the processing unit according to the relationship between the average processing volume score q and the processing volume score W2 of the processing unit that is not in the load state, it includes:

The control module is further used to obtain a processing volume score difference △Q between the processing volume score average score q and the processing volume score W2, △Q=q-W2, and the control module is further used to compare the processing volume score difference △Q with a preset processing volume score difference, and select a corresponding amount of data to be allocated to the processing unit according to the comparison result;

The control module is further used to pre-set a first preset processing volume score difference value △Q1 and a second preset processing volume score difference value △Q2, and the control module is further used to pre-set a first preset data volume C1, a second preset data volume C2 and a third preset data volume C3, and △Q1＜△Q2, C1＜C2＜C3;

When △Q≤△Q1, the control module selects the first preset data volume C1 to be distributed to the processing unit;

When △Q1＜△Q≤△Q2, the control module selects the second preset data volume C2 to be distributed to the processing unit;

When ΔQ>ΔQ2, the control module selects the third preset data volume C3 to be allocated to the processing unit.

Compared with the prior art, the beneficial effect of an artificial intelligence acceleration computing chip of the present invention is that: through the demand acquisition module, users can easily obtain and set preset demand conditions, establish a retrieval analysis formula, and provide guidance for subsequent data processing. Then, the crawler module intelligently retrieves the demand data according to the retrieval analysis formula, realizing the efficient acquisition of a huge data set. The data acquisition module is responsible for collecting the retrieved demand data and constructing the data set, providing a solid foundation for further processing. Secondly, in the processing module, multiple processing units are provided, so that the chip can perform multi-task parallel processing at the same time, significantly improving the computing efficiency. The scoring module evaluates the data processing capabilities of each processing unit and provides intelligent decision support for the data allocation of different task amounts. Through the control module, the chip can allocate the amount of data in the data set according to the score of the processing unit, realizing the reasonable allocation and efficient execution of tasks. In addition, the control module also obtains the average score of the processing volume score of the processing unit through formula-based processing, thereby more comprehensively evaluating the overall performance and providing an improvement for optimizing the use of computing resources.

BRIEF DESCRIPTION OF THE DRAWINGS

Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

FIG1 is a structural block diagram of an artificial intelligence acceleration computing chip provided in an embodiment of the present invention.

DETAILED DESCRIPTION

Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

In some embodiments of the present application, this embodiment provides an artificial intelligence acceleration computing chip, including: a demand acquisition module, a crawler module, a data acquisition module, a processing module, a scoring module and a control module. The demand acquisition module is used to obtain preset demand conditions and establish a retrieval analysis formula based on the preset demand conditions. The crawler module is electrically connected to the demand acquisition module, and the crawler module is used to retrieve demand data according to the retrieval analysis formula. The data acquisition module is electrically connected to the crawler module, and the data acquisition module is used to collect demand data and establish a data set based on the demand data. The processing module is provided with a number of processing units, and the processing module is used to perform task data processing on the demand data. The scoring module is used to obtain the data processing capabilities of a number of processing units, and to score the processing volume of the processing units according to the data processing capabilities. The control module is electrically connected to the processing module and the scoring module, respectively, and the control module is used to allocate the amount of data in the data set according to the processing volume scores of the number of processing units. Among them, the control module is also used to obtain the total processing volume score of the number of processing units, and obtain the average score of the processing volume score of the number of processing units based on the formula, and the formula is as follows:

q = Q/Y;

Among them, q is the average score of the processing capacity of several processing units, Q is the total score of the processing capacity of several processing units, and Y is the number of several processing units.

The control module is also used to allocate the data volume in the data set according to the relationship between the average processing volume score and the processing volume scores of several processing units.

It can be seen that through the demand acquisition module, the chip can flexibly obtain and establish preset demand conditions for the task, and then retrieve relevant demand data in real time through the crawler module. The data acquisition module is responsible for collecting and constructing data sets to provide rich materials for subsequent task processing. Several processing units are set in the processing module, which improve the overall processing efficiency through task data processing. The scoring module provides a basis for subsequent task allocation by evaluating the data processing capabilities of the processing unit. The control module plays a key role in the entire process. It allocates the amount of data in the data set according to the score of the processing unit, and calculates the average score of the processing volume of the processing unit through a formula, which provides a basis for the intelligent scheduling and parallel computing of the chip.

It is understandable that the overall computing efficiency can be effectively improved through intelligent task scheduling and data volume allocation. The scoring mechanism in the processing module helps to accurately evaluate the performance of the processing unit, thereby achieving more reasonable resource allocation. The formula calculation and relationship judgment in the control module enable the chip to automatically adapt to changes in the number of processing units, improving the chip's performance and efficiency in parallel computing and data processing of artificial intelligence big data, while greatly shortening the time for processing artificial intelligence big data.

In some embodiments of the present application, when the data acquisition module is used to collect demand data and establish a data set based on the demand data, the data acquisition module is also used to remove duplicate data from the demand data. The data acquisition module is also used to remove abnormal data from the demand data after deleting duplicate data, and the abnormal data includes: undisclosed decrypted data and incompletely disclosed data. The data acquisition module is also used to establish a data set from the demand data after removing duplicate data and abnormal data.

It can be seen that the demand data is first deduplicated through the acquisition module to ensure the cleanliness and efficiency of the data set. Secondly, the deduplicated data is subjected to abnormal data elimination, which involves eliminating undisclosed decrypted data and incompletely disclosed data, thereby improving the quality and availability of the data set. Finally, the demand data that has been deduplicated and processed for abnormal data is used to establish a data set, providing a clean and complete data foundation for subsequent processing and analysis.

It is understandable that through the multi-layer processing of the data acquisition module, the optimization and refinement of the required data are achieved. Deduplication processing ensures the efficient use of the data set and avoids the redundant impact of duplicate data on subsequent tasks. The elimination of abnormal data improves the accuracy of the data set and ensures the credibility and integrity of the data. The final data set is cleaner and more accurate, which helps to improve the accuracy and efficiency of the artificial intelligence acceleration computing chip in processing tasks, provides strong support for the overall data processing performance of the chip, and effectively avoids the great impact on the chip throughput when processing complex and repetitive data.

In some embodiments of the present application, when the scoring module scores the processing volume of the processing unit according to the data processing capability, it includes: the scoring module is also used to obtain the real-time task processing volume L of the processing unit, and the scoring module is also used to judge whether the real-time task processing volume of the processing unit is in a load state according to the relationship between the real-time task processing volume L and the preset load processing volume L0: when L≥L0, the control module judges that the real-time task processing volume of the processing unit is in a load state. When L＜L0, the control module judges that the real-time task processing volume of the processing unit is not in a load state. Among them, the scoring module is also used to obtain a processing unit that is not in a load state among a number of processing units, and to score the processing volume of the processing unit that is not in a load state according to the relationship between the real-time task processing volume L and the preset load processing volume L0.

In some embodiments of the present application, the scoring module is also used to obtain processing units that are not in a load state among several processing units, and to score the processing capacity of the processing units that are not in a load state according to the relationship between the real-time task processing capacity L and the preset load processing capacity L0, including: the scoring module is also used to obtain the processing capacity difference △L between the real-time task processing capacity L and the preset load processing capacity L0, △L=L-L0, the scoring module is also used to compare the processing capacity difference △L with the preset processing capacity difference, and select a relative scoring coefficient according to the comparison result to score the processing capacity of the processing units that are not in a load state: wherein the scoring module is also used to pre-set a first preset processing capacity difference △L1 and a second preset processing capacity difference △L2, and the scoring module is also used to pre-set a first preset scoring coefficient M1, a second preset scoring coefficient M2 and a third preset scoring coefficient M3, and △L1＜△L2, M1＜M2＜M3.

When ΔL≤ΔL1, the first preset scoring coefficient M1 is selected to score the processing capacity of the processing unit that is not in the load state.

When ΔL1＜ΔL≤ΔL2, the second preset scoring coefficient M2 is selected to score the processing capacity of the processing unit that is not in the load state.

When ΔL>ΔL2, the third preset scoring coefficient M3 is selected to score the processing capacity of the processing unit that is not in the load state.

When the scoring module selects the i-th preset scoring coefficient Mi to score the processing capacity of the processing unit not in the load state, I=1,2,3, and determines the processing capacity score of the processing unit not in the load state as W, and sets W1=M i.

It can be seen that, first, the scoring module obtains the real-time task processing volume of the processing unit, and determines whether the processing unit is in a load state by comparing it with the preset load processing volume. For processing units that are not in a load state, the scoring module further calculates the processing volume difference, and selects the corresponding scoring coefficient for the processing volume scoring of the processing unit according to the preset processing volume difference and the scoring coefficient. Specifically, by setting the preset processing volume difference and the corresponding scoring coefficient, the scoring module can dynamically adjust the selection of the scoring coefficient according to the relationship between the real-time task processing volume and the load processing volume of the processing unit, so as to achieve a more refined evaluation. When the processing volume of the processing unit deviates far from the load state, a higher scoring coefficient is selected, and vice versa, a lower scoring coefficient is selected, so that the score is closer to the actual operating state.

It is understandable that through the dynamic scoring mechanism, the scoring module can more accurately evaluate the real-time task processing capabilities of the processing unit, avoiding the shortcomings of relying solely on static scoring. Differentiated processing volume scoring enables the scoring module to more flexibly adapt to task requirements under different load conditions, improving the intelligent scheduling and resource utilization efficiency of the control module. This dynamic scoring mechanism is expected to achieve more accurate and efficient task processing in complex AI accelerated computing scenarios.

In some embodiments of the present application, when the scoring module selects the i-th preset scoring coefficient Mi to score the processing unit that is not in the load state, and determines that the processing volume score of the processing unit that is not in the load state is W, it includes: the scoring module is also used to obtain the real-time data processing speed K of the processing unit that is not in the load state, and judge whether the real-time data processing speed of the processing unit that is not in the load state meets the standard data processing speed according to the relationship between the real-time data processing speed K and the preset standard data processing speed K0: when K=K0, the scoring module judges that the real-time data processing speed of the processing unit meets the standard data processing speed, and does not adjust the processing volume score W of the processing unit that is not in the load state. When K≠K0, the scoring module judges that the real-time data processing speed of the processing unit does not meet the standard data processing speed, and adjusts the processing volume score W of the processing unit that is not in the load state according to the relationship between the real-time data processing speed K and the preset standard data processing speed K0.

In some embodiments of the present application, when the scoring module determines that the real-time data processing speed of the processing unit does not meet the standard data processing speed, and adjusts the processing volume score W of the processing unit that is not in the load state according to the relationship between the real-time data processing speed K and the preset standard data processing speed K0, the scoring module includes: the scoring module is also used to obtain the data processing speed difference △K between the real-time data processing speed K and the preset standard data processing speed K0, △K＝K-K0, the scoring module is also used to compare the processing speed difference △K with the preset processing speed difference, and select the corresponding adjustment coefficient according to the comparison result to adjust the processing volume score W of the processing unit that is not in the load state: wherein the scoring module is also used to pre-set the first preset processing speed difference △K1 and the second preset processing speed difference △K2. The scoring module is also used to pre-set the first preset adjustment coefficient N1, the second preset adjustment coefficient N2 and the third preset adjustment coefficient N3, and △K1＜△K2, N1＜0＜N2＜N3＜0.8.

When ΔK≤ΔK1, the third preset adjustment coefficient N3 is selected to adjust the processing capacity score W of the processing unit that is not in the load state.

When ΔK1＜ΔK≤0, the processing capacity score W of the processing unit that is not in the load state is not adjusted.

When 0＜△K≤△K2, the second preset adjustment coefficient N2 is selected to adjust the processing capacity score W of the processing unit that is not in the load state.

When ΔK>ΔK2, the first preset adjustment coefficient N1 is selected to adjust the processing capacity score W of the processing unit that is not in the load state.

When the scoring module selects the i-th preset adjustment coefficient Ni to adjust the processing volume score W of the processing unit not in the load state, i=1,2,3, and determines that the adjusted processing volume score of the processing unit not in the load state is W1, and sets W1=W*Ni.

It can be seen that, first, the scoring module obtains the real-time data processing speed of the processing unit that is not in a load state, and determines whether it meets the standard by comparing it with the preset standard data processing speed. According to the relationship between the real-time data processing speed and the standard speed, the scoring module can adjust the processing volume score of the processing unit that is not in a load state. Specifically, the scoring module dynamically determines whether the real-time data processing speed meets the standard by setting a preset processing speed difference and a corresponding adjustment coefficient, and adjusts the processing volume score of the processing unit accordingly. Different processing speed difference intervals use different adjustment coefficients, so that the scoring module and the control module can more flexibly evaluate and adjust the performance of the processing unit in different situations.

It is understandable that by dynamically adjusting the processing unit's processing volume score in real time, the scoring module can more accurately reflect its actual performance level. The evaluation and adjustment of the real-time data processing speed makes the scoring module and the control module more robust and able to adapt to data processing requirements under different load conditions. This dynamic adjustment mechanism is expected to improve the adaptability and performance stability of the overall chip for large data parallel processing, ensuring the best task processing effect in different workload scenarios.

In some embodiments of the present application, when the scoring module selects the i-th preset adjustment coefficient Ni to adjust the processing volume score W of the processing unit that is not in the load state, and determines that the adjusted processing volume score of the processing unit that is not in the load state is W1, it includes: the scoring module is also used to obtain the real-time read and write speed J of the processing unit that is not in the load state, and judge whether to adjust the adjusted processing volume score W1 of the processing unit that is not in the load state according to the relationship between the real-time read and write speed J and the preset standard read and write speed J0: when J=J0, the scoring module judges not to adjust the adjusted processing volume score W1 of the processing unit that is not in the load state. When J≠J0, the scoring module judges that the real-time read and write speed of the processing unit does not meet the standard read and write speed, and according to the relationship between the real-time read and write speed J and the preset standard read and write speed J0, the adjusted processing volume score W1 of the processing unit that is not in the load state is corrected.

In some embodiments of the present application, when the scoring module determines that the real-time read/write speed of the processing unit does not meet the standard read/write speed, and according to the relationship between the real-time read/write speed J and the preset standard read/write speed J0, the processing volume score W1 of the processing unit that is not in the load state after adjustment is corrected, including: the scoring module is also used to obtain the reading/writing speed difference △J between the real-time read/write speed J and the preset standard read/write speed J0, △J＝J-J0, the scoring module is also used to compare the reading/writing speed difference △J with the preset reading/writing speed difference, and select the corresponding correction coefficient according to the comparison result to correct the processing volume score W1 of the processing unit that is not in the load state after adjustment: wherein the scoring module is also used to pre-set the first preset reading/writing speed difference △J1 and the second preset reading/writing speed difference △J2. The scoring module is also used to pre-set the first preset correction coefficient B1, the second preset correction coefficient B2 and the third preset correction coefficient B3, and △J1＜△J2, B1＜0＜B2＜B3＜0.5.

When ΔJ≤ΔJ1, the third preset correction coefficient B3 is selected to correct the adjusted processing capacity score W1 of the processing unit that is not in the load state.

When ΔJ1<ΔJ≤0, the adjusted processing volume score W1 of the processing unit that is not in the load state is not corrected.

When 0＜△J≤△J2, the second preset correction coefficient B2 is selected to correct the adjusted processing capacity score W1 of the processing unit that is not in the load state.

When ΔJ>ΔJ2, the first preset correction coefficient B1 is selected to correct the adjusted processing capacity score W1 of the processing unit that is not in the load state.

When the scoring module selects the i-th preset correction coefficient Bi to correct the adjusted processing capacity score W1 of the processing unit not in the load state, i=1,2,3, and determines that the corrected processing capacity score of the processing unit not in the load state is W2, and sets W2=W1*N i.

It can be seen that, first, the scoring module obtains the real-time read and write speed of the processing unit, and determines whether the adjusted processing volume score needs to be corrected by comparing it with the preset standard read and write speed. According to the relationship between the real-time read and write speed and the standard speed, the scoring module can correct the processing volume score of the processing unit that is not in a load state. Specifically, the scoring module dynamically determines whether the real-time read and write speed meets the standard by setting a preset read and write speed difference and a corresponding correction coefficient, and corrects the adjusted processing volume score accordingly. Different read and write speed difference intervals use different correction coefficients, so that the scoring module can further scientifically evaluate the performance of the processing unit in different situations.

It is understandable that by dynamically correcting the real-time read and write speeds, the scoring module can more accurately reflect the actual performance level of the processing unit. Different correction coefficients enable the control module to more flexibly adapt to different read and write speed differences, further improving the overall chip's task allocation work when processing large amounts of data. This multi-level dynamic correction mechanism is expected to further improve the scoring module's performance evaluation of each processing unit, while enabling the control module to understand the performance of each processing unit, ensuring that the chip can achieve the best task processing effect under different data task processing load scenarios.

In some embodiments of the present application, the control module is also used to allocate the amount of data in the data set according to the relationship between the average score of the processing volume and the processing volume scores of several processing units, including: the control module is also used to obtain the processing volume scores W2 of several processing units that are not in a load state, and allocate the amount of data in the data set according to the relationship between the average score of the processing volume scores Q and the processing volume scores W2 of several processing units that are not in a load state: when W2≥Q, the control module allocates the amount of data in the data set to the processing unit according to the relationship between the average score of the processing volume scores Q and the processing volume scores W2 of the processing units that are not in a load state. When W2＜Q, the control module does not allocate the amount of data in the data set to the processing unit.

In some embodiments of the present application, when the control module allocates the data volume in the data set to the processing unit based on the relationship between the processing volume score average score Q and the processing volume score W2 of the processing unit that is not in a load state, it includes: the control module is also used to obtain the processing volume score difference △Q between the processing volume score average score Q and the processing volume score W2, △Q = Q-W2, the control module is also used to compare the processing volume score difference △Q with the preset processing volume score difference, and select the corresponding data volume to allocate to the processing unit based on the comparison result: wherein the control module is also used to pre-set a first preset processing volume score difference △Q1 and a second preset processing volume score difference △Q2, the control module is also used to pre-set a first preset data volume C1, a second preset data volume C2 and a third preset data volume C3, and △Q1＜△Q2, C1＜C2＜C3.

When ΔQ≤ΔQ1, the control module selects the first preset data volume C1 to be distributed to the processing unit.

When ΔQ1＜ΔQ≤ΔQ2, the control module selects the second preset data volume C2 to be distributed to the processing unit.

When ΔQ>ΔQ2, the control module selects the third preset data volume C3 to be distributed to the processing unit.

It can be seen that, first, the control module obtains the processing volume score of the processing unit that is not in the load state, and then determines whether to allocate the data volume in the data set to the processing unit based on the relationship between the average processing volume score and the processing volume score. By setting the preset processing volume score difference and the corresponding data volume allocation preset value, the control module can flexibly adjust the allocation in the data set according to the dynamic relationship between the processing volume score average and the processing volume score of the processing unit that is not in the load state. Specifically, the control module calculates the processing volume score difference and sets different data volume allocation thresholds based on the preset processing volume score difference. Based on the comparison result between the processing volume score difference and the preset difference, the control module selects the corresponding data volume allocation preset value to ensure that the data volume in the data set can be reasonably and dynamically allocated according to the real-time performance situation.

It is understandable that by dynamically allocating the amount of data in the data set, the control module can more flexibly allocate tasks according to the real-time performance of the processing unit to achieve load balancing and resource optimization. According to the real-time processing capabilities of different processing units, ensure that the data in the data set is effectively allocated to the processing unit with better performance, and improve the processing efficiency of the overall chip when processing big data. This dynamic allocation architecture mechanism is expected to optimize task scheduling in artificial intelligence accelerated computing scenarios, improve chip hardware performance, and adapt to the needs of different load conditions.

In the above embodiment, through the demand acquisition module, the user can easily obtain and set the preset demand conditions, establish a retrieval analysis formula, and provide guidance for subsequent data processing. Then, the crawler module intelligently retrieves the demand data according to the retrieval analysis formula, realizing the efficient acquisition of a huge data set. The data acquisition module is responsible for collecting the retrieved demand data and constructing the data set, providing a solid foundation for further processing. Secondly, in the processing module, multiple processing units are provided, so that the chip can perform multi-task parallel processing at the same time, significantly improving the computing efficiency. The scoring module evaluates the data processing capabilities of each processing unit, providing intelligent decision support for the data allocation of different task amounts. Through the control module, the chip can allocate the amount of data in the data set according to the score of the processing unit, realizing the reasonable allocation and efficient execution of tasks. In addition, the control module also obtains the average score of the processing volume score of the processing unit through formula-based processing, thereby more comprehensively evaluating the overall performance and providing an improvement for optimizing the use of computing resources.

Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

The present application is described with reference to the flowchart and/or block diagram of the method, device (system) and computer program commodity according to the embodiment of the present application. It should be understood that each process and/or box in the flowchart and/or block diagram, as well as the combination of the process and/or box in the flowchart and/or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the function specified in one process or multiple processes in the flowchart and/or one box or multiple boxes in the block diagram.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims (10)

1. An artificial intelligence acceleration operation chip, which is characterized by comprising:

the demand acquisition module is used for acquiring preset demand conditions and establishing a retrieval analysis formula according to the preset demand conditions;

The crawler module is electrically connected with the demand acquisition module and is used for searching the demand data according to the search analysis type;

The data acquisition module is electrically connected with the crawler module and is used for acquiring the demand data and establishing a data set according to the demand data;

the processing module is provided with a plurality of processing units and is used for performing task data processing on the demand data;

the scoring module is used for acquiring the data processing capacity of the processing units and scoring the processing capacity of the processing units according to the data processing capacity;

The control module is respectively and electrically connected with the processing module and the scoring module and is used for scoring and distributing the data volume in the data set according to the processing volume of the processing units; wherein,

The control module is further used for obtaining total evaluation scores of the processing amounts of the processing units, and obtaining average scores of the processing amounts of the processing units according to a formula, wherein the formula is as follows:

q＝Q/Y；

Wherein Q is the average score of the processing capacity of a plurality of processing units, Q is the total score of the processing capacity of a plurality of processing units, and Y is the number of a plurality of processing units;

the control module is further configured to allocate data amounts in the data set according to a relationship between the average of the throughput scores and the throughput scores of the plurality of processing units.

2. The artificial intelligence acceleration operation chip of claim 1, wherein the data acquisition module is configured to acquire the demand data and to create a data set according to the demand data, and comprises:

the data acquisition module is also used for removing repeated data in the required data;

the data acquisition module is further configured to reject abnormal data in the requirement data for removing the repeated data, where the abnormal data includes: unrevealed decrypted data and incompletely disclosed data;

The data acquisition module is also used for establishing a data set from the demand data after the repeated data and the abnormal data are removed.

3. The artificial intelligence acceleration operation chip of claim 2, wherein the scoring module, when scoring the processing unit in terms of the data processing capacity, comprises:

The scoring module is further used for acquiring the real-time task processing amount L of the processing unit, and judging that the real-time task processing amount of the processing unit is in a load state according to the relation between the real-time task processing amount L and a preset load processing amount L0;

When L is more than or equal to L0, the control module judges that the real-time task processing capacity of the processing unit is in a load state;

When L is less than L0, the control module judges that the real-time task processing capacity of the processing unit is not in a load state;

The scoring module is further configured to obtain processing units that are not in a load state from the plurality of processing units, and score the processing units that are not in the load state according to a relationship between the real-time task processing amount L and a preset load processing amount L0.

4. The artificial intelligence acceleration operation chip of claim 3, wherein the scoring module is further configured to obtain a processing unit that is not in a loaded state from among the plurality of processing units, and perform a processing amount scoring on the processing unit that is not in a loaded state according to a relationship between the real-time task processing amount L and a preset load processing amount L0, where the processing unit includes:

The scoring module is further used for obtaining a processing amount difference delta L between the real-time task processing amount L and a preset load processing amount L0, wherein delta L=L-L0, comparing the processing amount difference delta L with the preset processing amount difference, and scoring the processing unit which is not in a load state according to a scoring coefficient which is selected to be opposite according to a comparison result;

The scoring module is further used for presetting a first preset scoring coefficient M1, a second preset scoring coefficient M2 and a third preset scoring coefficient M3, wherein the delta L1 is less than delta L2, and M1 is less than M2 and less than M3;

When DeltaL is less than or equal to DeltaL 1, selecting the first preset scoring coefficient M1 to score the processing amount of the processing unit which is not in the load state;

When DeltaL 1 < DeltaL2 is less than or equal to DeltaL 2, selecting the second preset scoring coefficient M2 to score the processing amount of the processing unit which is not in the load state;

when DeltaL > DeltaL2, selecting the third preset scoring coefficient M3 to score the processing capacity of the processing unit which is not in the load state;

When the scoring module selects the I preset scoring coefficient Mi to score the processing unit which is not in the load state, I=1, 2 and 3, and determines the processing amount score of the processing unit which is not in the load state as W, and setting W1=Mi.

5. The artificial intelligence acceleration operation chip according to claim 4, wherein when the scoring module selects the i-th preset scoring coefficient Mi to score the processing units not in the loaded state and determines that the processing amount of the processing units not in the loaded state is scored as W, the method comprises:

The scoring module is further used for acquiring the real-time data processing speed K of the processing unit which is not in a load state, and judging whether the real-time data processing speed of the processing unit which is not in the load state accords with the standard data processing speed according to the relation between the real-time data processing speed K and the preset standard data processing speed K0;

When k=k0, the scoring module determines that the real-time data processing speed of the processing unit meets a standard data processing speed, and does not adjust the throughput score W of the processing unit not in the load state;

When K is not equal to K0, the scoring module judges that the real-time data processing speed of the processing unit does not accord with the standard data processing speed, and adjusts the processing capacity score W of the processing unit which is not in a load state according to the relation between the real-time data processing speed K and the preset standard data processing speed K0.

6. The artificial intelligence acceleration operation chip of claim 5, wherein the scoring module determines that the real-time data processing speed of the processing unit does not meet the standard data processing speed, and adjusts the throughput score W of the processing unit not under load according to the relationship between the real-time data processing speed K and the preset standard data processing speed K0, including:

The scoring module is further configured to obtain a data processing speed difference Δk between the real-time data processing speed K and a preset standard data processing speed K0, where Δk=k-K0, compare the processing speed difference Δk with the preset processing speed difference, and select a corresponding adjustment coefficient according to a comparison result to adjust a processing capacity score W of the processing unit that is not in a load state:

The scoring module is further used for presetting a first preset processing speed difference delta K1 and a second preset processing speed difference delta K2; the scoring module is also used for presetting a first preset adjustment coefficient N1, a second preset adjustment coefficient N2 and a third preset adjustment coefficient N3, wherein DeltaK 1 < DeltaK2, N1 is more than 0 and less than N2 and N3 is more than 0.8;

When delta K is less than or equal to delta K1, selecting the third preset adjustment coefficient N3 to adjust the throughput score W of the processing unit which is not in a load state;

when DeltaK 1 < DeltaKis less than or equal to 0, the processing capacity score W of the processing unit which is not in a load state is not adjusted;

when the delta K is less than or equal to 0 and less than or equal to delta K2, selecting the second preset adjustment coefficient N2 to adjust the throughput score W of the processing unit which is not in a load state;

When DeltaK > DeltaK2, selecting the first preset adjustment coefficient N1 to adjust the processing capacity score W of the processing unit which is not in a load state;

When the scoring module selects the i preset adjustment coefficient Ni to adjust the processing capacity score W of the processing unit not in the load state, i=1, 2,3, and determines that the adjusted processing capacity score of the processing unit not in the load state is W1, and setting w1=w×ni.

7. The artificial intelligence acceleration operation chip according to claim 6, wherein when the scoring module selects the i-th preset adjustment coefficient Ni to adjust the processing capacity score W of the processing unit not in the loaded state and determines that the adjusted processing capacity score of the processing unit not in the loaded state is W1, the method comprises:

The scoring module is further used for acquiring the real-time read-write speed J of the processing unit which is not in the load state, and judging whether to adjust the adjusted throughput score W1 of the processing unit which is not in the load state according to the relation between the real-time read-write speed J and the preset standard read-write speed J0;

when j=j0, the scoring module determines that the adjusted throughput score W1 of the processing unit not in the load state is not adjusted;

When J is not equal to J0, the scoring module judges that the real-time read-write speed of the processing unit does not accord with the standard read-write speed, and corrects the adjusted processing capacity score W1 of the processing unit which is not in a load state according to the relation between the real-time read-write speed J and the preset standard read-write speed J0.

8. The artificial intelligence acceleration operation chip of claim 7, wherein the scoring module determines that the real-time read-write speed of the processing unit does not meet the standard read-write speed, and corrects the adjusted throughput score W1 of the processing unit not under load according to the relationship between the real-time read-write speed J and the preset standard read-write speed J0, including:

The scoring module is further configured to obtain a difference Δj between the real-time read-write speed J and a preset standard read-write speed J0, where Δj=j-J0, compare the difference Δj with the preset read-write speed difference, and select a corresponding correction coefficient according to the comparison result to correct the adjusted throughput score W1 of the processing unit that is not in the loaded state:

The scoring module is further used for presetting a first preset read-write speed difference delta J1 and a second preset read-write speed difference delta J2; the scoring module is also used for presetting a first preset correction coefficient B1, a second preset correction coefficient B2 and a third preset correction coefficient B3, wherein delta J1 < [ delta ] J2, B1 is more than 0 and less than 0.5, and B2 is more than 0 and less than 0.5;

When delta J is less than or equal to delta J1, selecting the third preset correction coefficient B3 to correct the adjusted throughput score W1 of the processing unit which is not in the load state;

when DeltaJ 1 < DeltaJis less than or equal to 0, not correcting the adjusted processing capacity score W1 of the processing unit which is not in the load state;

When delta J is less than or equal to 0 and less than delta J2, selecting the second preset correction coefficient B2 to correct the adjusted throughput score W1 of the processing unit which is not in the load state;

When DeltaJ > DeltaJ2, selecting the first preset correction coefficient B1 to correct the adjusted processing capacity score W1 of the processing unit which is not in the load state;

When the scoring module selects the i preset correction coefficient Bi to correct the adjusted processing capacity score W1 of the processing unit which is not in the load state, i=1, 2,3, and determines that the corrected processing capacity score of the processing unit which is not in the load state is W2, and sets w2=w1×ni.

9. The artificial intelligence acceleration operation chip of claim 8, wherein the control module is further configured to, when allocating the data volume in the data set according to a relationship between the average of the throughput scores and the throughput scores of the plurality of processing units, include:

The control module is further used for obtaining the processing capacity scores W2 of the processing units which are not in the load state, and distributing the data capacity in the data set according to the relation between the average q of the processing capacity scores and the processing capacity scores W2 of the processing units which are not in the load state;

When W2 is more than or equal to q, the control module distributes the data volume in the data set to the processing unit according to the relation between the average score q of the processing volume and the score W2 of the processing unit which is not in a load state;

When W2 < q, the control module does not allocate the amount of data in the dataset to the processing unit.

10. The artificial intelligence acceleration operation chip of claim 9, wherein the control module assigns the data amount in the data set to the processing unit according to a relationship between the average q of the processing amount scores and the processing amount score W2 of the processing unit not under load, and comprises:

the control module is further used for obtaining a throughput score difference DeltaQ between the throughput score average Q and the throughput score W2, deltaq=Q-W2, comparing the throughput score difference DeltaQ with a preset throughput score difference, and selecting corresponding data according to a comparison result and distributing the data to the processing unit;

The control module is further used for presetting a first preset processing amount scoring difference delta Q1 and a second preset processing amount scoring difference delta Q2, and presetting a first preset data amount C1, a second preset data amount C2 and a third preset data amount C3, wherein delta Q1 < [ delta ] Q2, and C1 < C2 < C3;

When DeltaQ is less than or equal to DeltaQ 1, the control module selects the first preset data quantity C1 to be distributed to the processing unit;

when DeltaQ 1 < DeltaQis less than or equal to DeltaQ 2, the control module selects the second preset data quantity C2 to be distributed to the processing unit;

when DeltaQ > DeltaQ2, the control module selects the third preset data quantity C3 to be distributed to the processing unit.