CN113722085B - Graphics resource allocation method and allocation system - Google Patents

Abstract

The invention provides a graphic resource distribution method and a graphic resource distribution system. The method comprises the following steps: obtaining first association information between a first graphic processing unit and at least one central processing unit in the plurality of graphic processing units; obtaining second association information between the first graphics processing unit and at least one second graphics processing unit of the plurality of graphics processing units; establishing a grouping and rubbing structure of the plurality of graphic processing units according to the first association information and the second association information; and executing the graphic resource allocation of the graphic processing units according to the grouping and rubbing structure. Therefore, the image processing efficiency can be effectively improved.

Description

Graphics resource allocation method and allocation system

Technical Field

The present invention relates to a system resource allocation technology, and in particular to a graphic resource allocation method and allocation system.

Background technique

With the advancement of technology, the requirements for the computing speed of computer systems are getting higher and higher. Therefore, some computer systems can be configured with multiple Graphics Processing Units (GPUs) to assist the Central Processing Unit (CPU) in performing graphics-related computing. However, in the existing working mechanism, the graphics processing work related to a certain central processing unit is often randomly assigned to one or more graphics processing units for processing, without considering whether the communication path (or signal transmission path) between the graphics processing unit that receives the graphics processing work and the central processing unit is too complicated, thereby reducing system performance.

Summary of the invention

The present invention provides a method and system for allocating graphic resources, which can effectively improve the efficiency of graphic processing.

An embodiment of the present invention provides a method for allocating graphics resources, which includes: obtaining first association information between a first graphics processing unit among multiple graphics processing units and at least one central processing unit; obtaining second association information between the first graphics processing unit and at least one second graphics processing unit among the multiple graphics processing units; establishing a cluster topology structure of the multiple graphics processing units based on the first association information and the second association information; and executing graphics resource allocation of the multiple graphics processing units based on the cluster topology structure.

An embodiment of the present invention further provides a system for allocating graphics resources, which includes at least one central processing unit, multiple graphics processing units, a storage device, and a graphics resource allocator. The multiple graphics processing units are connected to the at least one central processing unit. The storage device is connected to the at least one central processing unit and the multiple graphics processing units. The graphics resource allocator is connected to the storage device. The graphics resource allocator is used to obtain first association information between a first graphics processing unit among the multiple graphics processing units and the at least one central processing unit from the storage device. The graphics resource allocator is further used to obtain second association information between the first graphics processing unit and at least one second graphics processing unit among the multiple graphics processing units from the storage device. The graphics resource allocator is further used to establish a cluster topology structure of the multiple graphics processing units based on the first association information and the second association information. The graphics resource allocator is further used to perform graphics resource allocation of the multiple graphics processing units based on the cluster topology structure.

Based on the above, according to the first association information between the first graphics processing unit and the central processing unit and the second association information between the first graphics processing unit and at least one second graphics processing unit, a cluster topology structure of multiple graphics processing units can be established. Thereafter, according to the cluster topology structure, the graphics resources of the multiple graphics processing units can be properly allocated, thereby effectively improving the graphics processing efficiency.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram of a system for allocating graphics resources according to an embodiment of the present invention;

2 is a schematic diagram showing a physical connection relationship between a central processing unit and a graphics processing unit according to an embodiment of the present invention;

FIG3 is a schematic diagram of first association information according to an embodiment of the present invention;

FIG4 is a schematic diagram of second association information according to an embodiment of the present invention;

FIG5 is a schematic diagram of a cluster topology structure according to an embodiment of the present invention;

FIG. 6 is a flow chart of a method for allocating graphics resources according to an embodiment of the present invention.

Detailed ways

Reference will now be made in detail to exemplary embodiments of the present invention, examples of which are illustrated in the accompanying drawings. Whenever possible, the same reference numerals are used in the drawings and the description to refer to the same or like parts.

FIG1 is a schematic diagram of a system for allocating graphics resources according to an embodiment of the present invention. Referring to FIG1 , the system (also referred to as a system for allocating graphics resources) 10 can be implemented in various electronic devices (or computer devices) having data processing and graphics processing functions, such as notebook computers, desktop computers, industrial computers, or servers.

The system 10 includes a central processing unit (CPU) 11 (1) to 11 (n), a graphics processing unit (GPU) 12 (1) to 12 (m), a storage device 13, and a graphics resource allocator 14. Each of the central processing units (CPU) 11 (1) to 11 (n) can refer to a central processing unit chip. Each of the graphics processing units 12 (1) to 12 (m) can refer to a graphics processing unit chip. In operation, the central processing units 11 (1) to 11 (n) and the graphics processing units 12 (1) to 12 (m) can work together to jointly complete data processing and graphics processing tasks. The total number of the central processing units 11 (1) to 11 (n) and the total number of the graphics processing units 12 (1) to 12 (m) can be determined by practical needs and the present invention is not limited thereto.

The storage device 13 is used to store data. The storage device 13 may include a volatile memory module and a non-volatile memory module. The volatile memory module may include a memory medium such as a random access memory (RAM) that can store data volatilely. The non-volatile memory module may include a memory medium such as a read-only memory (ROM) and/or a flash memory that can store data non-volatilely.

The graphics resource allocator 14 is connected to the CPUs 11(1)-11(n), the graphics processing units 12(1)-12(m) and the storage device 13. In one embodiment, the graphics resource allocator 14 can be implemented as a programmable general-purpose or special-purpose microprocessor, a digital signal processor (DSP), a programmable controller, an application specific integrated circuit (ASIC), a programmable logic device (PLD) or other similar devices or a combination of these devices. In one embodiment, the graphics resource allocator 14 can also be implemented as a program code and stored in the storage device 13.

The storage device 13 stores table information 131 and 132. The table information 131 and 132 are automatically generated by, for example, scanning related hardware information by the operating system (e.g., Windows operating system) or system tools such as BIOS of the system 10 after the central processing units 11(1)-11(n) and the graphics processing units 12(1)-12(m) are installed on a motherboard (not shown) of the system 10.

The table information 131 records the association information between the central processing units 11(1)-11(n) and the graphics processing units 12(1)-12(m). In particular, the table information 131 may record the association information (also referred to as the first association information) between a certain graphics processing unit (also referred to as the first graphics processing unit) among the graphics processing units 12(1)-12(m) and the central processing units 11(1)-11(n). For example, the first association information may include the core number information of a certain central processing unit (also referred to as the first central processing unit) that is preset and preferentially used by the first graphics processing unit among the central processing units 11(1)-11(n).

The table information 132 records the association information between the GPUs 12(1)-12(m). In particular, the table information 132 may record the association information (also referred to as second association information) between the first GPU and the remaining GPUs (also referred to as second GPUs) in the GPUs 12(1)-12(m). For example, the second association information may include the proximity level information between the first GPU and the second GPU.

The graphics resource allocator 14 can read the table information 131 and 132 from the storage device 13 to obtain the first association information and the second association information. Based on the first association information and the second association information, the graphics resource allocator 14 can establish a cluster topology structure of the graphics processing units 12 (1) to 12 (m). Thereafter, the graphics resource allocator 14 can perform graphics resource allocation of the graphics processing units 12 (1) to 12 (m) according to the established cluster topology structure. For example, when it is necessary to execute a graphics processing task related to at least one of the central processing units 11 (1) to 11 (n), the graphics resource allocator 14 can select a suitable graphics processing unit from the graphics processing units 12 (1) to 12 (m) according to the established cluster topology structure and allocate the graphics processing task to the selected graphics processing unit for processing.

FIG2 is a schematic diagram showing the physical connection relationship between a central processing unit and a graphics processing unit according to an embodiment of the present invention. Referring to FIG1 and FIG2 , it is assumed that the central processing units 11(1) to 11(n) include central processing units 11(1) and 11(2) and the graphics processing units 12(1) to 12(m) include graphics processing units 12(1) to 12(8).

The central processing units 11(1) and 11(2) are connected to the graphics processing units 12(1) to 12(8) via interfaces 21(1) to 21(4). For example, the central processing unit 11(1) is connected to the graphics processing units 12(1) and 12(2) via interface 21(1) and is connected to the graphics processing units 12(3) and 12(4) via interface 21(2). The central processing unit 11(2) is connected to the graphics processing units 12(5) and 12(6) via interface 21(3) and is connected to the graphics processing units 12(7) and 12(8) via interface 21(4). The interfaces 21(1) to 21(4) may respectively include PCIe switches or other types of physical connection interfaces.

It should be noted that in the embodiment of FIG. 2 , when the upper-level central processing units 11(1) and 11(2) assign graphics processing tasks to the lower-level graphics processing units 12(1) to 12(8) or when the upper-level central processing units 11(1) and 11(2) receive processing results of graphics processing tasks from the lower-level graphics processing units 12(1) to 12(8), the central processing unit 11(1) can communicate directly with the graphics processing units 12(1) to 12(4) via interfaces 21(1) and 21(2), and the central processing unit 11(2) can communicate directly with the graphics processing units 12(5) to 12(8) via interfaces 21(3) and 21(4). However, when it comes to the exchange of tasks and/or processing results across central processing units, central processing unit 11(1) must communicate with graphics processing units 12(5) to 12(8) via central processing unit 11(2), and central processing unit 11(2) must communicate with graphics processing units 12(1) to 12(4) via central processing unit 11(1).

Based on the above-mentioned communication limitations, if a graphics processing unit is randomly selected to assist a specific central processing unit in performing graphics processing, the signal transmission may be delayed due to poor selection of the graphics processing unit, thereby causing a decrease in the overall graphics processing performance. In one embodiment, the established cluster topology structure can at least partially reflect the physical connection relationship between the upper-level central processing unit and the lower-level graphics processing unit. Therefore, when it is necessary to allocate graphics processing resources, the graphics resource allocator 14 can select a more suitable one or more graphics processing units according to this cluster topology structure to assist a specific central processing unit in performing graphics operations, thereby improving the overall graphics processing performance.

For the sake of convenience, it is assumed that the central processing units 11(1) and 11(1) are numbered CPU#1 and CPU#2, respectively. In addition, it is assumed that the graphics processing units 12(1) to 12(8) are numbered GPU#1, GPU#2, GPU#4, GPU#5, GPU#3, GPU#6, GPU#7, and GPU#8, respectively.

FIG. 3 is a schematic diagram of first association information according to an embodiment of the present invention. FIG. 4 is a schematic diagram of second association information according to an embodiment of the present invention. Referring to FIGS. 1 to 3 , it is assumed that the first association information includes association information 31. Association information 31 may reflect the association between central processing units 11(1) to 11(2) and graphics processing units 12(1) to 12(8). For example, association information 31 may reflect that the core numbers of the central processing units that are preset for priority use by graphics processing units GPU#1 to GPU#8 are CPU core#1, CPU core#1, CPU core#2, CPU core#1, CPU core#1, CPU core#2, CPU core#2, and CPU core#2, respectively. Among them, CPU core#1 is, for example, the core number of the central processing unit 11(1), and CPU core#2 is, for example, the core number of the central processing unit 11(2).

According to the association information 31, the graphics resource allocator 14 can obtain the group to which each graphics processing unit belongs. For example, according to the association information 31, the graphics resource allocator 14 can determine that the graphics processing units GPU#1 to GPU#8 belong to group 1, group 1, group 2, group 1, group 1, group 2, group 2, and group 2, respectively. The graphics processing units belonging to group 1 may include at least some of the graphics processing units that are preset to give priority to the use of the central processing unit CPU#1 (for example, graphics processing units GPU#1, GPU#2, GPU#4, GPU#5). The graphics processing units belonging to group 2 may include at least some of the graphics processing units that are preset to give priority to the use of the central processing unit CPU#2 (for example, graphics processing units GPU#3, GPU#6, GPU#7, GPU#8).

1 to 4, it is assumed that the second association information includes association information 41. The association information 41 may reflect the association between the GPUs 12(1) to 12(8). For example, the association information 41 may reflect the proximity level information between a specific GPU (i.e., the first GPU) and the remaining GPUs (i.e., the second GPU).

In one embodiment, the proximity rank information may be represented by Rank 1 where the distances between the multiple GPUs are relatively close and Rank 2 where the distances between the multiple GPUs are relatively far. For example, the association information 41 may record that the distances between GPU#1 and GPU#2, GPU#3, GPU#4 and GPU#5 are relatively close, and the distances between GPU#1 and GPU#6, GPU#7 and GPU#8 are relatively far; the distances between GPU#2 and GPU#1, GPU#4 and GPU#5 are relatively close, and the distances between GPU#3, GPU#6, GPU#7 and GPU#8 are relatively far; the distances between GPU#3 and GPU#1, GPU#6, GPU#7 and GPU#8 are relatively close, and the distances between GPU#2, GPU#4 and GPU#5 are relatively far, etc.

It should be noted that the information recorded in the association information 41 may not fully and correctly reflect the actual physical connection relationship between the upper-level central processing units 11(1) and 11(2) and the lower-level graphics processing units 12(1) to 12(8). For example, for graphics processing unit GPU#1, the association information 41 reflects that graphics processing unit GPU#3 is a graphics processing unit that is closer, but in fact, graphics processing unit GPU#3 is connected to another central processing unit 11(2), not central processing unit 11(1). Therefore, if the cluster topology structure is established based solely on the association information 41, misjudgment may occur, thereby affecting the subsequent graphics processing performance.

Therefore, in one embodiment, when establishing the cluster topology structure, the graphics resource allocator 14 may select the graphics processing units (also referred to as the third graphics processing units) that belong to the same group and are closer to each graphics processing unit. For example, according to the association information 41, the graphics resource allocator 14 may select the graphics processing units GPU#2, GPU#3, GPU#4 and GPU#5 as the graphics processing units that are closest to the graphics processing unit GPU#1, select the graphics processing units GPU#1, GPU#4 and GPU#5 as the graphics processing units that are closest to the graphics processing unit GPU#2, and select the graphics processing units GPU#1, GPU#6, GPU#7 and GPU#8 as the graphics processing units that are closest to the graphics processing unit GPU#3. Then, according to whether the selected third graphics processing unit and the corresponding first graphics processing unit belong to the same group, the graphics resource allocator 14 may determine the cluster topology structure related to the first graphics processing unit in the finally established cluster topology structure (also referred to as the first cluster topology structure).

FIG5 is a schematic diagram of a cluster topology structure according to an embodiment of the present invention. Referring to FIGS. 1 to 5, based on the association information 31 and 41, when establishing the cluster topology structure 51 of the GPU#1 belonging to group 1, the GPU#2, GPU#4 and GPU#5 belonging to the same group 1 and being the closest to the GPU#1 will be retained in the cluster topology structure 51, while the GPU#3 belonging to a different group 2 but being mistakenly recorded as also being the closest to the GPU#1 will be removed from the cluster topology structure 51.

When establishing the cluster topology structure 52 of the GPU#2 belonging to group 1, the GPU#1, GPU#4 and GPU#5 belonging to the same group 1 and being the closest GPU#2 are retained in the cluster topology structure 52. When establishing the cluster topology structure 53 of the GPU#3 belonging to group 2, the GPU#6, GPU#7 and GPU#8 belonging to the same group 2 and being the closest GPU#3 are retained in the cluster topology structure 53, while the GPU#1 belonging to a different group 1 but being mistakenly recorded as also being the closest GPU#3 is removed from the cluster topology structure 53. Similarly, the cluster topology structures corresponding to the remaining GPU#4 to GPU#8 can also be established.

Thereafter, when it is desired to allocate graphics resources required for graphics computing, the graphics resource allocator 14 may select the graphics processing unit GPU#1 and at least one of the graphics processing units GPU#2, GPU#4 and GPU#5 belonging to the cluster topology structure 51 to collaborate in graphics computing, or select the graphics processing unit GPU#3 and at least one of the graphics processing units GPU#6, GPU#7 and GPU#8 belonging to the cluster topology structure 53 to collaborate in graphics computing, etc. Compared with randomly selecting available graphics processing units for graphics processing, selecting graphics processing units according to the cluster topology structure established in the aforementioned embodiment can effectively improve the overall graphics computing performance.

It should be noted that, although FIG. 2 is used as an example of the physical connection relationship between the upper-layer central processing unit and the lower-layer graphics processing unit in the aforementioned embodiment, the present invention is not limited thereto. In another embodiment, the physical connection relationship between the upper-layer central processing unit and the lower-layer graphics processing unit may also have other connection forms, such as more layers or via more or fewer interfaces, etc., which are not limited by the present invention. In addition, no matter what the physical connection relationship between the upper-layer central processing unit and the lower-layer graphics processing unit is, once the central processing unit and the graphics processing unit are installed on the motherboard, the table information 131 and 132 of FIG. 1 can be automatically generated. Thereafter, the corresponding cluster topology structure can be established based on the table information 131 and 132, which will not be repeated here.

FIG6 is a flow chart of a method for allocating graphics resources according to an embodiment of the present invention. Referring to FIG6, in step S601, first association information between a first graphics processing unit among a plurality of graphics processing units and at least one central processing unit is obtained. In step S602, second association information between the first graphics processing unit and at least one second graphics processing unit among the plurality of graphics processing units is obtained. In step S603, a cluster topology structure of the plurality of graphics processing units is established according to the first association information and the second association information. In step S604, graphics resource allocation of the plurality of graphics processing units is performed according to the cluster topology structure.

However, each step in FIG6 has been described in detail above, and will not be repeated here. It is worth noting that each step in FIG6 can be implemented as multiple program codes or circuits, and the present invention is not limited thereto. In addition, the method of FIG6 can be used in conjunction with the above exemplary embodiments, or can be used alone, and the present invention is not limited thereto.

In summary, the embodiment of the present invention can group the graphics processing units according to the first association information. Then, according to the grouping result and the second association information, a grouping topology structure of the graphics processing units can be established. According to the established grouping topology structure, the graphics resources of multiple graphics processing units can be properly allocated and used, thereby effectively improving the graphics processing efficiency in the future.

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 aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims (6)

1. A method for allocating graphics resources, comprising:

Obtaining first association information between a first graphic processing unit and at least one central processing unit in a plurality of graphic processing units, wherein the first association information reflects that a central processing unit preset for use by the first graphic processing unit is a first central processing unit in the at least one central processing unit;

Obtaining second association information between the first graphics processing unit and at least one second graphics processing unit of the plurality of graphics processing units, wherein the second association information reflects a distance between the first graphics processing unit and the at least one second graphics processing unit;

establishing a grouping and rubbing structure of the plurality of graphic processing units according to the first association information and the second association information; and

Executing the graphic resource allocation of the plurality of graphic processing units according to the grouping topology,

The step of establishing the grouping topology structure of the plurality of graphic processing units according to the first association information and the second association information comprises the following steps:

obtaining a first group to which the first graphic processing unit belongs according to the first association information;

selecting at least one of the at least one second graphic processing unit as a third graphic processing unit nearest to the first graphic processing unit according to the second association information; and

Determining a first group topology related to the first graphic processing unit in the group topology according to whether the third graphic processing unit belongs to the first group,

Wherein determining the first group topology related to the first graphic processing unit in the group topology according to whether the third graphic processing unit belongs to the first group comprises:

If the third graphic processing unit belongs to the first group, reserving the third graphic processing unit in the first grouping roller structure; and

If the third GPU does not belong to the first group, removing the third GPU from the first split-cluster-topology,

Wherein the step of executing the graphic resource allocation of the plurality of graphic processing units according to the group topology comprises:

And selecting at least one graphic processing unit belonging to the first grouping and cleaning structure to cooperatively perform graphic operation.

2. The allocation method of graphic resources according to claim 1, wherein the first association information records core number information of the first central processing unit, and the core number information includes a core number of the first central processing unit.

3. The method of allocating graphics resources according to claim 1, wherein the second association information records proximity level information between the first graphics processing unit and the at least one second graphics processing unit, and the proximity level information reflects the distance between the first graphics processing unit and the at least one second graphics processing unit.

4. A graphics resource allocation system, comprising:

at least one central processing unit;

a plurality of graphic processing units connected to the at least one central processing unit;

A storage device; and

A graphics resource allocator coupled to said at least one central processing unit, said storage device, and said plurality of graphics processing units,

Wherein the graphics resource allocator is configured to obtain first association information between a first graphics processing unit of the plurality of graphics processing units and the at least one central processing unit from the storage device, wherein the first association information reflects that a central processing unit preset for use by the first graphics processing unit is a first central processing unit of the at least one central processing unit,

The graphics resource allocator is further configured to obtain second association information between the first graphics processing unit and at least one second graphics processing unit of the plurality of graphics processing units from the storage device, wherein the second association information reflects a distance between the first graphics processing unit and the at least one second graphics processing unit,

The graphic resource allocator is further configured to establish a packet topology of the plurality of graphic processing units according to the first association information and the second association information, and

The graphic resource allocator is further configured to perform the graphic resource allocation of the plurality of graphic processing units according to the grouping topology,

The operation of establishing the grouping topology structure of the plurality of graphic processing units according to the first association information and the second association information comprises the following steps:

obtaining a first group to which the first graphic processing unit belongs according to the first association information;

selecting at least one of the at least one second graphic processing unit as a third graphic processing unit nearest to the first graphic processing unit according to the second association information; and

Determining a first group topology related to the first graphic processing unit in the group topology according to whether the third graphic processing unit belongs to the first group,

Wherein determining the first group topology related to the first graphic processing unit in the group topology according to whether the third graphic processing unit belongs to the first group comprises:

If the third graphic processing unit belongs to the first group, reserving the third graphic processing unit in the first grouping roller structure; and

If the third GPU does not belong to the first group, removing the third GPU from the first split-cluster-topology,

Wherein performing the graphic resource allocation of the plurality of graphic processing units according to the grouping topology comprises:

And selecting at least one graphic processing unit belonging to the first grouping and cleaning structure to cooperatively perform graphic operation.

5. The graphic resource allocation system according to claim 4, wherein the first association information records core number information of the first central processing unit, and the core number information includes a core number of the first central processing unit.

6. The graphics resource allocation system according to claim 4, wherein said second association information records proximity level information between said first graphics processing unit and said at least one second graphics processing unit, and said proximity level information reflects said distance between said first graphics processing unit and said at least one second graphics processing unit.