[go: up one dir, main page]

TWI698833B - Graphics processor system - Google Patents

Graphics processor system Download PDF

Info

Publication number
TWI698833B
TWI698833B TW107143644A TW107143644A TWI698833B TW I698833 B TWI698833 B TW I698833B TW 107143644 A TW107143644 A TW 107143644A TW 107143644 A TW107143644 A TW 107143644A TW I698833 B TWI698833 B TW I698833B
Authority
TW
Taiwan
Prior art keywords
speed serial
serial computer
computer expansion
expansion bus
bus standard
Prior art date
Application number
TW107143644A
Other languages
Chinese (zh)
Other versions
TW202022793A (en
Inventor
許繼委
Original Assignee
英業達股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 英業達股份有限公司 filed Critical 英業達股份有限公司
Priority to TW107143644A priority Critical patent/TWI698833B/en
Publication of TW202022793A publication Critical patent/TW202022793A/en
Application granted granted Critical
Publication of TWI698833B publication Critical patent/TWI698833B/en

Links

Images

Landscapes

  • Multi Processors (AREA)

Abstract

The present invention provides a graphics processor system, the graphics processor system comprising: a central processing unit, a plurality of graphics processors communicatively coupled to the central processing unit, a bus communication protocol switch, and the plurality of graphics processor is connected to implement mutual communication between the graphics processors, a management board is separately connected to the bus communication protocol exchange for managing the bus communication protocol switch. The invention provides a bus bar communication protocol switch, which can maximize and equalize the peer-to-peer network communication bandwidth between the graphics processors, and the theoretical maximum bidirectional bandwidth can reach 300 GB/s, and can further expand an eight-graphic processing. The system forms a sixteen graphics processor system that enables peer-to-peer network communication of any two graphics processors.

Description

圖形處理器系統Graphics processor system

本發明屬於資料處理技術領域,涉及一種資料處理系統,特別是涉及一種圖形處理器系統。 The invention belongs to the technical field of data processing, and relates to a data processing system, in particular to a graphics processor system.

隨著虛擬實境和人工智慧技術的快速發展,對具有海量資料計算和處理能力的系統的需求逐漸攀升。圖形處理器(Graphics Processing Unit,GPU)正是以高計算性能著稱,近年來顯得格外受歡迎。 With the rapid development of virtual reality and artificial intelligence technologies, the demand for systems with massive data computing and processing capabilities has gradually increased. Graphics Processing Unit (GPU) is known for its high computing performance and has become extremely popular in recent years.

GPU又稱顯示核心、視覺處理器、顯示晶片,其是一種專門在個人電腦、工作站、遊戲機和一些移動設備(如平板電腦、智慧手機等)上執行圖像運算工作的微處理器。GPU的用途是將電腦系統所需要的顯示資訊進行轉換驅動,並向顯示器提供行掃描信號,控制顯示器的正確顯示,是連接顯示器和個人電腦主機板的重要元件,也是“人機對話”的重要設備之一。 GPU, also known as display core, visual processor, and display chip, is a microprocessor that performs image operations on personal computers, workstations, game consoles, and some mobile devices (such as tablet computers, smart phones, etc.). The purpose of GPU is to convert and drive the display information required by the computer system, and to provide line scan signals to the display to control the correct display of the display. It is an important component for connecting the display to the motherboard of a personal computer, and it is also an important part of "human-machine dialogue". One of the equipment.

在實際應用中,採用多GPU系統來增強圖形處理能力是常用手段,但由於GPU與CPU之間通信介面的限制,以及GPU與GPU之間通信頻寬的限制,多GPU系統的圖形處理能力也是受限的,並不能隨著GPU數量的增多而呈現指數增強的效果。 In practical applications, the use of multi-GPU systems to enhance graphics processing capabilities is a common method. However, due to the limitations of the communication interface between GPU and CPU and the limitation of the communication bandwidth between GPUs and GPUs, the graphics processing capabilities of multi-GPU systems are also Limited, and can not show the effect of exponential enhancement as the number of GPUs increases.

鑒於以上所述現有技術的缺點,本發明的目的在於提供一種圖形處理器系統,用於解決現有多GPU系統的通信速率低的問題。 In view of the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide a graphics processor system for solving the problem of low communication rate of the existing multi-GPU system.

為實現上述目的及其他相關目的,本發明提供一種圖形處理器系統,所述圖形處理器系統包括:一中央處理器;複數個圖形處理器,與所述中央處理器通信相連;一匯流排通信協定交換機,與所述複數個圖形處理器相連,用於實現各圖形處理器之間的相互通信;一管理板,與所述匯流排通信協定交換機分別通信相連,用於對所述匯流排通信協定交換機進行管理。 In order to achieve the above-mentioned objects and other related objects, the present invention provides a graphics processor system. The graphics processor system includes: a central processing unit; a plurality of graphics processing units connected in communication with the central processing unit; and a bus communication A protocol switch, connected to the plurality of graphics processors, and used to implement mutual communication between the graphics processors; a management board, respectively, communicatively connected to the bus communication protocol switch, and used to communicate with the bus Agreement switch for management.

於本發明的一實施例中,所述匯流排通信協定交換機包括:至少一匯流排通信協定交換模組;所述匯流排通信協定交換模組包括6個匯流排通信協定交換模塊;各匯流排通信協定交換模組設有16組匯流排通信協定埠;各匯流排通信協定交換模組的每組匯流排通信協定埠用於連接一圖形處理器或擴展;匯流排通信協定管理模組,與所述各匯流排通信協定交換模組相連,用於管理各匯流排通信協定交換模組的各組匯流排通信協定埠之間的通信。 In an embodiment of the present invention, the bus communication protocol switch includes: at least one bus communication protocol switching module; the bus communication protocol switching module includes 6 bus communication protocol switching modules; each bus The communication protocol exchange module is equipped with 16 sets of bus communication protocol ports; each set of bus communication protocol ports of each bus communication protocol exchange module is used to connect a graphics processor or expansion; the bus communication protocol management module, and The bus communication protocol exchange modules are connected, and are used to manage the communication between each group of bus communication protocol ports of each bus communication protocol exchange module.

於本發明的一實施例中,所述匯流排通信協定交換模組中的8組匯流排通信協定埠用於一一對應連接8個圖形處理器,所述各匯流排通信協定交換模組中的剩餘8組匯流排通信協定埠用於擴展。 In an embodiment of the present invention, the 8 sets of bus communication protocol ports in the bus communication protocol switching module are used to connect 8 graphics processors in a one-to-one correspondence, and each bus communication protocol switching module is The remaining 8 sets of bus communication protocol ports are used for expansion.

於本發明的一實施例中,所述圖形處理器系統還包括:一高速串列電腦擴展匯流排標準交換機組,與所述中央處理器和複數個圖形處理器分別相連,用於擴展所述中央處理器的高速串列電腦擴展匯流排標準埠,實現所述圖形處理器與所述中央處理器之間的相互通信;一高速串列電腦擴展匯流排標準擴展模組,與所述高速串列電腦擴展匯流排標準交換機組、所述匯流排通信協定 管理模組和所述管理板分別通信相連;所述管理板通過所述高速串列電腦擴展匯流排標準擴展模組對所述高速串列電腦擴展匯流排標準交換機組和所述匯流排通信協定交換機進行管理。 In an embodiment of the present invention, the graphics processor system further includes: a high-speed serial computer expansion bus standard switch group, which is connected to the central processing unit and a plurality of graphics processing units, respectively, for expanding the The high-speed serial computer expansion bus standard port of the central processing unit realizes the mutual communication between the graphics processor and the central processing unit; a high-speed serial computer expansion bus standard expansion module is connected to the high-speed serial computer Column computer extended bus standard switch group, said bus communication protocol The management module and the management board are respectively connected in communication; the management board expands the bus standard switch group and the bus communication protocol to the high-speed serial computer through the high-speed serial computer expansion bus standard expansion module The switch is managed.

於本發明的一實施例中,所述高速串列電腦擴展匯流排標準交換機組包括:一第一高速串列電腦擴展匯流排標準交換機,與所述中央處理器相連;一第二高速串列電腦擴展匯流排標準交換機,與所述第一高速串列電腦擴展匯流排標準交換機相連;所述第二高速串列電腦擴展匯流排標準交換機下行連接複數個圖形處理器;一第三高速串列電腦擴展匯流排標準交換機,與所述第一高速串列電腦擴展匯流排標準交換機相連;所述第三高速串列電腦擴展匯流排標準交換機下行連接複數個圖形處理器;所述管理板與所述第一高速串列電腦擴展匯流排標準交換機、第二高速串列電腦擴展匯流排標準交換機和第三高速串列電腦擴展匯流排標準交換機分別相連,用於對所述第一高速串列電腦擴展匯流排標準交換機、第二高速串列電腦擴展匯流排標準交換機和第三高速串列電腦擴展匯流排標準交換機進行管理設置。 In an embodiment of the present invention, the high-speed serial computer expansion bus standard switch group includes: a first high-speed serial computer expansion bus standard switch connected to the central processing unit; and a second high-speed serial computer expansion bus standard switch The computer expansion bus standard switch is connected to the first high-speed serial computer expansion bus standard switch; the second high-speed serial computer expansion bus standard switch is downstream connected to a plurality of graphics processors; a third high-speed serial The computer expansion bus standard switch is connected to the first high-speed serial computer expansion bus standard switch; the third high-speed serial computer expansion bus standard switch is connected downstream to a plurality of graphics processors; the management board is connected to the The first high-speed serial computer expansion bus standard switch, the second high-speed serial computer expansion bus standard switch, and the third high-speed serial computer expansion bus standard switch are respectively connected to the first high-speed serial computer The expansion bus standard switch, the second high-speed serial computer expansion bus standard switch and the third high-speed serial computer expansion bus standard switch for management settings.

於本發明的一實施例中,所述第一高速串列電腦擴展匯流排標準交換機包括:至少一第一主機埠,用於與所述中央處理器通信相連;至少2個第一光纖通信埠,用於分別與所述第二高速串列電腦擴展匯流排標準交換機和所述第三高速串列電腦擴展匯流排標準交換機通信相連;至少一第一上行埠,用於與所述管理板通信相連。 In an embodiment of the present invention, the first high-speed serial computer expansion bus standard switch includes: at least one first host port for communicating with the central processing unit; at least two first optical fiber communication ports , Used to respectively communicate with the second high-speed serial computer expansion bus standard switch and the third high-speed serial computer expansion bus standard switch; at least one first uplink port is used to communicate with the management board Connected.

於本發明的一實施例中,所述第二高速串列電腦擴展匯流排標準交換機包括:至少一第二光纖通信埠,用於與所述第一高速串列電腦擴展匯流排標準交換機的一第一光纖通信埠通信相連;至少一第二上行埠,用於與所述管 理板通信相連;至少一第二下行埠,用於與圖形處理器通信相連。 In an embodiment of the present invention, the second high-speed serial computer expansion bus standard switch includes: at least one second optical fiber communication port for communicating with one of the first high-speed serial computer expansion bus standard switch The first optical fiber communication port is communicatively connected; at least one second uplink port is used to communicate with the tube The management board is connected in communication; at least one second downstream port is used to communicate with the graphics processor.

於本發明的一實施例中,所述第三高速串列電腦擴展匯流排標準交換機包括:至少一第三光纖通信埠,用於與所述第一高速串列電腦擴展匯流排標準交換機的另一第一光纖通信埠通信相連;至少一第三上行埠,用於與所述管理板通信相連;至少一第三下行埠,用於與圖形處理器通信相連。 In an embodiment of the present invention, the third high-speed serial computer expansion bus standard switch includes: at least one third optical fiber communication port for connecting to another standard switch of the first high-speed serial computer expansion bus A first optical fiber communication port is used for communication connection; at least one third uplink port is used for communication connection with the management board; at least one third downlink port is used for communication connection with a graphics processor.

於本發明的一實施例中,所述第二高速串列電腦擴展匯流排標準交換機的一第二下行埠用於連接網卡或固態硬碟;或/和所述第三高速串列電腦擴展匯流排標準交換機的一第三下行埠用於連接網卡或固態硬碟。 In an embodiment of the present invention, a second downstream port of the second high-speed serial computer expansion bus standard switch is used to connect a network card or a solid state drive; or/and the third high-speed serial computer expansion bus A third downstream port of the standard switch is used to connect a network card or a solid state drive.

於本發明的一實施例中,所述第一高速串列電腦擴展匯流排標準交換機、第二高速串列電腦擴展匯流排標準交換機和第三高速串列電腦擴展匯流排標準交換機均工作於虛擬模式;所述管理板對所述第一高速串列電腦擴展匯流排標準交換機、第二高速串列電腦擴展匯流排標準交換機和第三高速串列電腦擴展匯流排標準交換機進行動態管理。 In an embodiment of the present invention, the first high-speed serial computer expansion bus standard switch, the second high-speed serial computer expansion bus standard switch, and the third high-speed serial computer expansion bus standard switch all work in virtual Mode; The management board dynamically manages the first high-speed serial computer expansion bus standard switch, the second high-speed serial computer expansion bus standard switch and the third high-speed serial computer expansion bus standard switch.

如上所述,本發明所述的圖形處理器系統,具有以下有益效果: As mentioned above, the graphics processor system of the present invention has the following beneficial effects:

本發明提出了匯流排通信協定交換機,其可將圖形處理器間的對等網路通信頻寬最大化和均衡化,理論最大雙向頻寬可達300GB/s,還可以再擴展一個八圖形處理器系統,構成一個十六圖形處理器系統,可實現任意兩圖形處理器的對等網路通信。 The present invention proposes a bus communication protocol switch, which can maximize and equalize the communication bandwidth of the peer-to-peer network between graphics processors. The theoretical maximum two-way bandwidth can reach 300GB/s, and it can also expand an eight graphics processing The processor system constitutes a sixteen graphics processor system, which can realize peer-to-peer network communication between any two graphics processors.

100:圖形處理器系統 100: Graphics processor system

110:中央處理器 110: central processing unit

120:圖形處理器 120: graphics processor

130:匯流排通信協定交換機 130: Bus communication protocol switch

131:匯流排通信協定交換模組 131: Bus communication protocol exchange module

1311:匯流排通信協定交換模組 1311: Bus communication protocol exchange module

132:匯流排通信協定管理模組 132: Bus communication protocol management module

140:管理板 140: Management Board

150:高速串列電腦擴展匯流排標準交換機組 150: High-speed serial computer expansion bus standard switch group

151:第一高速串列電腦擴展匯流排標準交換機 151: The first high-speed serial computer expansion bus standard switch

1511:第一主機埠 1511: First host port

1512:第一光纖通信埠 1512: The first optical fiber communication port

1513:第一上行埠 1513: first uplink port

152:第二高速串列電腦擴展匯流排標準交換機 152: The second high-speed serial computer expansion bus standard switch

1521:第二光纖通信埠 1521: Second fiber optic communication port

1522:第二上行埠 1522: second upstream port

1523:第二下行埠 1523: second downstream port

153:第三高速串列電腦擴展匯流排標準交換機 153: The third high-speed serial computer expansion bus standard switch

1531:第三光纖通信埠 1531: Third fiber optic communication port

1532:第三上行埠 1532: third upstream port

1533:第三下行埠 1533: third downstream port

160:高速串列電腦擴展匯流排標準擴展模組 160: High-speed serial computer expansion bus standard expansion module

圖1顯示為本發明實施例所述的圖形處理器系統的一種示例性實現結構示意圖。 FIG. 1 shows a schematic diagram of an exemplary implementation structure of a graphics processor system according to an embodiment of the present invention.

圖2顯示為本發明實施例所述的圖形處理器系統的匯流排通信協定交換機的一種示例性實現結構示意圖。 Fig. 2 is a schematic diagram showing an exemplary implementation structure of the bus communication protocol switch of the graphics processor system according to the embodiment of the present invention.

圖3顯示為本發明實施例所述的圖形處理器系統的匯流排通信協定交換機的一種示例性通信結構示意圖。 FIG. 3 is a schematic diagram of an exemplary communication structure of the bus communication protocol switch of the graphics processor system according to the embodiment of the present invention.

圖4顯示為本發明實施例所述的圖形處理器系統的另一種示例性實現結構示意圖。 FIG. 4 is a schematic diagram showing another exemplary implementation structure of the graphics processor system according to the embodiment of the present invention.

圖5A顯示為本發明實施例所述的圖形處理器系統的第一高速串列電腦擴展匯流排標準交換機的一種示例性埠結構示意圖。 FIG. 5A is a schematic diagram showing an exemplary port structure of the first high-speed serial computer expansion bus standard switch of the graphics processor system according to an embodiment of the present invention.

圖5B顯示為本發明實施例所述的圖形處理器系統的第二高速串列電腦擴展匯流排標準交換機的一種示例性埠結構示意圖。 5B is a schematic diagram showing an exemplary port structure of the second high-speed serial computer expansion bus standard switch of the graphics processor system according to the embodiment of the present invention.

圖5C顯示為本發明實施例所述的圖形處理器系統的第三高速串列電腦擴展匯流排標準交換機的一種示例性埠結構示意圖。 5C is a schematic diagram showing an exemplary port structure of the third high-speed serial computer expansion bus standard switch of the graphics processor system according to the embodiment of the present invention.

圖6A和6B顯示為本發明實施例所述的圖形處理器系統的一種具體實施結構示意圖。 6A and 6B are schematic diagrams showing a specific implementation structure of a graphics processor system according to an embodiment of the present invention.

圖7顯示為本發明實施例所述的圖形處理器系統的另一種具體實施結構示意圖。 FIG. 7 is a schematic diagram of another specific implementation structure of the graphics processor system according to the embodiment of the present invention.

以下通過特定的具體實例說明本發明的實施方式,本領域技術人員可由本說明書所揭露的內容輕易地瞭解本發明的其他優點與功效。本發明還可以通過另外不同的具體實施方式加以實施或應用,本說明書中的各項細節也可以基於不同觀點與應用,在沒有背離本發明的精神下進行各種修飾或改變。 需說明的是,在不衝突的情況下,以下實施例及實施例中的特徵可以相互組合。 The following describes the implementation of the present invention through specific specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and the features in the embodiments can be combined with each other if there is no conflict.

需要說明的是,以下實施例中所提供的圖示僅以示意方式說明本發明的基本構想,遂圖式中僅顯示與本發明中有關的組件而非按照實際實施時的元件數目、形狀及尺寸繪製,其實際實施時各元件的型態、數量及比例可為一種隨意的改變,且其元件佈局型態也可能更為複雜。 It should be noted that the illustrations provided in the following embodiments only illustrate the basic idea of the present invention in a schematic manner. The figures only show the components related to the present invention instead of the number, shape, and number of components in actual implementation. For the size drawing, the type, number, and ratio of each component can be changed at will during actual implementation, and the component layout type may be more complicated.

多圖形處理器(Graphics Processing Unit,GPU)系統除了對GPU與中央處理器(Central Processing Unit Processor,CPU)之間有高頻寬性能有需求之外,還要求GPU之間有很高的頻寬能力,即對等網路(Peer to Peer,點對點)頻寬。為了提高多GPU系統的頻寬,本發明提供一種圖形處理器系統,如圖1所示,所述圖形處理器系統100包括:一中央處理器(CPU)110,複數個圖形處理器120,一匯流排通信協定交換機130,或/和一管理板140。所述複數個圖形處理器120與所述中央處理器110通信相連。所述匯流排通信協定交換機130與所述複數個圖形處理器120相連,用於實現各圖形處理器之間的相互通信。所述管理板140與所述匯流排通信協定交換機130分別通信相連,用於對所述匯流排通信協定交換機進行管理。 A multi-graphics processing unit (GPU) system not only requires high bandwidth performance between the GPU and the central processing unit (CPU), but also requires high bandwidth capabilities between the GPUs. That is, the Peer to Peer (Peer to Peer) bandwidth. In order to increase the bandwidth of a multi-GPU system, the present invention provides a graphics processor system. As shown in FIG. 1, the graphics processor system 100 includes: a central processing unit (CPU) 110, a plurality of graphics processors 120, The bus communication protocol switch 130, or/and a management board 140. The plurality of graphics processors 120 are connected to the central processing unit 110 in communication. The bus communication protocol switch 130 is connected to the plurality of graphics processors 120, and is used to implement mutual communication between the graphics processors. The management board 140 is respectively communicatively connected with the bus communication protocol switch 130, and is used to manage the bus communication protocol switch.

本發明中,NVLink是英偉達(NVIDIA)開發並推出的一種匯流排及其通信協定。NVLink採用點對點結構、串列傳輸,用於GPU與GPU之間的連接,也可用於CPU與GPU之間的連接。NVLink埠是GPU與GPU之間或者CPU與GPU之間的點對點通訊連接埠。所述匯流排通信協定交換機130即為NVLink交換機。 In the present invention, NVLink is a bus and its communication protocol developed and launched by NVIDIA. NVLink adopts a point-to-point structure and serial transmission, which is used for the connection between GPU and GPU, and can also be used for connection between CPU and GPU. The NVLink port is a point-to-point communication port between GPU and GPU or between CPU and GPU. The bus communication protocol switch 130 is an NVLink switch.

參見圖2所示,於本發明一實施例中,所述匯流排通信協定交換機130包括:至少一匯流排通信協定交換模組131,匯流排通信協定管理模組 132;所述匯流排通信協定交換模組131包括6個匯流排通信協定交換模塊1311;各匯流排通信協定交換模塊1311設有16組匯流排通信協定埠;各匯流排通信協定交換模組的每組匯流排通信協定埠用於連接一圖形處理器或擴展;所述匯流排通信協定管理模組132與所述各匯流排通信協定交換模塊1311相連,用於管理各匯流排通信協定交換模組的各組匯流排通信協定埠之間的通信。 Referring to FIG. 2, in an embodiment of the present invention, the bus communication protocol switch 130 includes: at least one bus communication protocol switching module 131, and a bus communication protocol management module 132; The bus communication protocol exchange module 131 includes 6 bus communication protocol exchange modules 1311; each bus communication protocol exchange module 1311 is provided with 16 sets of bus communication protocol ports; each bus communication protocol exchange module Each set of bus communication protocol ports is used to connect a graphics processor or an extension; the bus communication protocol management module 132 is connected to each bus communication protocol exchange module 1311 for managing each bus communication protocol exchange module Communication between the bus communication protocol ports of each group in the group.

參見圖3所示,於本發明一實施例中,所述匯流排通信協定交換模組1311中的8組匯流排通信協定埠用於一一對應連接8個圖形處理器,所述各匯流排通信協定交換模組中的剩餘8組匯流排通信協定埠用於擴展。 Referring to FIG. 3, in an embodiment of the present invention, the 8 sets of bus communication protocol ports in the bus communication protocol switch module 1311 are used to connect 8 graphics processors in a one-to-one correspondence, and each bus The remaining 8 sets of bus communication protocol ports in the communication protocol exchange module are used for expansion.

參見圖4所示,於本發明一實施例中,所述圖形處理器系統100還包括:一高速串列電腦擴展匯流排標準交換機組150,或/和一高速串列電腦擴展匯流排標準擴展模組160。所述高速串列電腦擴展匯流排標準交換機組150與所述中央處理器110和複數個圖形處理器120分別相連,用於擴展所述中央處理器的高速串列電腦擴展匯流排標準埠,實現所述圖形處理器與所述中央處理器之間的相互通信。所述高速串列電腦擴展匯流排標準擴展模組160與所述高速串列電腦擴展匯流排標準交換機組150、所述匯流排通信協定管理模組132和所述管理板140分別通信相連。述管理板140通過所述高速串列電腦擴展匯流排標準擴展模組160對所述高速串列電腦擴展匯流排標準交換機組150和所述匯流排通信協定交換機130進行管理。 Referring to FIG. 4, in an embodiment of the present invention, the graphics processor system 100 further includes: a high-speed serial computer expansion bus standard switch group 150, or/and a high-speed serial computer expansion bus standard expansion Module 160. The high-speed serial computer expansion bus standard switch group 150 is respectively connected to the central processing unit 110 and a plurality of graphics processors 120, and is used to expand the high-speed serial computer expansion bus standard port of the central processing unit to realize The mutual communication between the graphics processor and the central processing unit. The high-speed serial computer expansion bus standard expansion module 160 is respectively communicatively connected with the high-speed serial computer expansion bus standard switch group 150, the bus communication protocol management module 132 and the management board 140. The management board 140 manages the high-speed serial computer expansion bus standard switch group 150 and the bus communication protocol switch 130 through the high-speed serial computer expansion bus standard expansion module 160.

參見圖4所示,於本發明一實施例中,所述高速串列電腦擴展匯流排標準交換機組150包括:一第一高速串列電腦擴展匯流排標準交換機151,一第二高速串列電腦擴展匯流排標準交換機152,或/和一第三高速串列電腦擴展匯流排標準交換機153。所述第一高速串列電腦擴展匯流排標準交換機151與 所述中央處理器110相連。所述第二高速串列電腦擴展匯流排標準交換機152與所述第一高速串列電腦擴展匯流排標準交換機151相連;所述第二高速串列電腦擴展匯流排標準交換機152下行連接複數個圖形處理器120。所述第三高速串列電腦擴展匯流排標準交換機153與所述第一高速串列電腦擴展匯流排標準交換機151相連;所述第三高速串列電腦擴展匯流排標準交換機153下行連接複數個圖形處理器120。所述管理板140與所述第一高速串列電腦擴展匯流排標準交換機151、第二高速串列電腦擴展匯流排標準交換機152和第三高速串列電腦擴展匯流排標準交換機153分別相連,用於對所述第一高速串列電腦擴展匯流排標準交換機151、第二高速串列電腦擴展匯流排標準交換機152和第三高速串列電腦擴展匯流排標準交換機153進行管理設置。 Referring to FIG. 4, in an embodiment of the present invention, the high-speed serial computer expansion bus standard switch group 150 includes: a first high-speed serial computer expansion bus standard switch 151, and a second high-speed serial computer An extended bus standard switch 152, or/and a third high-speed serial computer extended bus standard switch 153. The first high-speed serial computer expansion bus standard switch 151 and The central processing unit 110 is connected. The second high-speed serial computer expansion bus standard switch 152 is connected to the first high-speed serial computer expansion bus standard switch 151; the second high-speed serial computer expansion bus standard switch 152 is downstream connected to a plurality of graphics The processor 120. The third high-speed serial computer expansion bus standard switch 153 is connected to the first high-speed serial computer expansion bus standard switch 151; the third high-speed serial computer expansion bus standard switch 153 is connected downstream with a plurality of graphics The processor 120. The management board 140 is connected to the first high-speed serial computer expansion bus standard switch 151, the second high-speed serial computer expansion bus standard switch 152, and the third high-speed serial computer expansion bus standard switch 153, respectively. The first high-speed serial computer expansion bus standard switch 151, the second high-speed serial computer expansion bus standard switch 152, and the third high-speed serial computer expansion bus standard switch 153 are managed.

進一步,所述第一高速串列電腦擴展匯流排標準交換機151、第二高速串列電腦擴展匯流排標準交換機152和第三高速串列電腦擴展匯流排標準交換機153均工作於虛擬模式;所述管理板140對所述第一高速串列電腦擴展匯流排標準交換機151、第二高速串列電腦擴展匯流排標準交換機152和第三高速串列電腦擴展匯流排標準交換機153進行動態管理。 Further, the first high-speed serial computer expansion bus standard switch 151, the second high-speed serial computer expansion bus standard switch 152, and the third high-speed serial computer expansion bus standard switch 153 all work in a virtual mode; The management board 140 dynamically manages the first high-speed serial computer expansion bus standard switch 151, the second high-speed serial computer expansion bus standard switch 152, and the third high-speed serial computer expansion bus standard switch 153.

參見圖5A所示,於本發明一實施例中,所述第一高速串列電腦擴展匯流排標準交換機151包括:至少一第一主機埠1511,至少2個第一光纖通信埠1512,至少一第一上行埠1513。所述至少一第一主機埠1511用於與所述中央處理器通信相連;所述至少2個第一光纖通信埠1512用於分別與所述第二高速串列電腦擴展匯流排標準交換機152和所述第三高速串列電腦擴展匯流排標準交換機153通信相連;所述至少一第一上行埠1513用於與所述管理板通信相連。 Referring to FIG. 5A, in an embodiment of the present invention, the first high-speed serial computer expansion bus standard switch 151 includes: at least one first host port 1511, at least two first optical fiber communication ports 1512, and at least one The first uplink port 1513. The at least one first host port 1511 is used to communicate with the central processing unit; the at least two first optical fiber communication ports 1512 are used to communicate with the second high-speed serial computer expansion bus standard switch 152 and The third high-speed serial computer expansion bus standard switch 153 is communicatively connected; the at least one first uplink port 1513 is used to communicatively connect to the management board.

參見圖5B所示,於本發明一實施例中,所述第二高速串列電腦擴展匯流排標準交換機152包括:至少一第二光纖通信埠1521,至少一第二上行埠1522,至少一第二下行埠1523。所述至少一第二光纖通信埠1521用於與所述第一高速串列電腦擴展匯流排標準交換機的一第一光纖通信埠1512通信相連;所述至少一第二上行埠1522用於與所述管理板140通信相連;所述至少一第二下行埠1523用於與圖形處理器120通信相連。所述第二高速串列電腦擴展匯流排標準交換機152的一第二下行埠用於連接網卡或固態硬碟。 Referring to FIG. 5B, in an embodiment of the present invention, the second high-speed serial computer expansion bus standard switch 152 includes: at least one second optical fiber communication port 1521, at least one second uplink port 1522, and at least one second The second downstream port is 1523. The at least one second optical fiber communication port 1521 is used to communicate with a first optical fiber communication port 1512 of the first high-speed serial computer expansion bus standard switch; the at least one second uplink port 1522 is used to communicate with all The management board 140 is communicatively connected; the at least one second downlink port 1523 is used for communicatively connected to the graphics processor 120. A second downstream port of the second high-speed serial computer expansion bus standard switch 152 is used to connect a network card or a solid state drive.

參見圖5C所示,於本發明一實施例中,所述第三高速串列電腦擴展匯流排標準交換機153包括:至少一第三光纖通信埠1531,至少一第三上行埠1532,至少一第三下行埠1533。所述至少一第三光纖通信埠1531用於與所述第一高速串列電腦擴展匯流排標準交換機的另一第一光纖通信埠通信相連;所述至少一第三上行埠1532用於與所述管理板通信相連;所述至少一第三下行埠1533用於與圖形處理器通信相連。所述第三高速串列電腦擴展匯流排標準交換機153的一第三下行埠用於連接網卡或固態硬碟。 Referring to FIG. 5C, in an embodiment of the present invention, the third high-speed serial computer expansion bus standard switch 153 includes: at least one third optical fiber communication port 1531, at least one third uplink port 1532, and at least one Three downstream ports 1533. The at least one third optical fiber communication port 1531 is used to communicate with another first optical fiber communication port of the first high-speed serial computer expansion bus standard switch; the at least one third uplink port 1532 is used to communicate with all The management board is communicatively connected; the at least one third downstream port 1533 is used for communicative connection with the graphics processor. A third downstream port of the third high-speed serial computer expansion bus standard switch 153 is used to connect a network card or a solid state drive.

隨著時代的進步,客戶對GPU系統性能的需求不斷提升,與此同時,GPU商家的技術也一直在進步,而我們需要設計出優良的架構把GPU產品的高性能充分發揮出來,以便提供給客戶的產品能勝任相應的業務。 With the progress of the times, customers’ demands for GPU system performance continue to increase. At the same time, the technology of GPU vendors has also been improving, and we need to design an excellent architecture to give full play to the high performance of GPU products in order to provide The customer's product is competent for the corresponding business.

原SXM2 GPU系統採用三個交換機(switch)的架構,用來擴展CPU的PCIE port(PCIE埠)給GPU。其中一級交換機(switch)採用虛擬模式(virtual mode),用來支援雙路X16 PCIE到CPU,給CPU和GPU的通信提供充足的頻寬。在本發明所述的新一代SXM3 GPU系統中依然可以採用;對於GPU之間的相互通信,利用NVLink互連提供部分GPU之間的直接對等網路(Peer to Peer)通信,另外一部分則利用PCIE switch(PCIE交換機)的光纖通信mode(光纖通信模式)來實現,如圖6A和6B所示。然而,這部分GPU對等網路的通信依然採用的是PCIE協定,最高速度為8Gbps,遠低於NVLink互連的速度25Gbps。因此,在SXM3 GPU系統中,利用匯流排通信協定交換機(NVLink switch)可以實現所有GPU間的通信都採用NVLink協議,這樣可以最大化任意GPU間的對等網路頻寬,系統拓撲如圖7所示。 The original SXM2 GPU system uses a three-switch architecture to extend the PCIE port of the CPU to the GPU. Among them, the first-level switch adopts virtual mode to support dual X16 PCIE to the CPU, providing sufficient bandwidth for the communication between the CPU and GPU. It can still be used in the new generation SXM3 GPU system of the present invention; for the mutual communication between GPUs, NVLink interconnection is used to provide a direct peer-to-peer network (Peer to Peer) communication, the other part is realized by the optical fiber communication mode of the PCIE switch (PCIE switch), as shown in Figures 6A and 6B. However, the communication of this part of the GPU peer-to-peer network still uses the PCIE protocol, with a maximum speed of 8Gbps, which is much lower than the 25Gbps interconnection speed of NVLink. Therefore, in the SXM3 GPU system, the use of a bus communication protocol switch (NVLink switch) can realize that all communications between GPUs use the NVLink protocol, which can maximize the peer-to-peer network bandwidth between any GPUs. The system topology is shown in Figure 7. Shown.

圖7所示的SXM3 GPU系統的拓撲圖中,CPU和GPU通信的部分依然採用三個PCIE switch(PEX9797)的結構,第一個switch為虛擬模式(virtual mode)工作模式,內部分為兩個虛擬橋VS0和VS1,兩個虛擬橋分別接一組X16 PCIE到CPU,實現GPU和CPU之間32個PCIE lane的通信頻寬;由於NVLink光纖通信需要一組X4 PCIE用作NV-Switch的管理,第二個switch外接一個PEX8749作為PCIE的擴展,同樣的工作在virtual mode,其中一個虛擬橋VS0用作NVLink光纖通信的管理和擴展一組X16 PCIE接網卡以及SSD等;另外一個虛擬橋VS1用作PCIE switch光纖通信的管理,下行分別接三個PEX9797的管理埠,為X1 PCIE,上行通過HD mini SAS接管理板,為X4 PCIE,預留一下行控制埠(control port)用作PCIE switch光纖通信擴展,為X1 PCIE。GPU和GPU間的對等網路通信採用的是NVLink光纖通信結構。每個SXM3 GPU有六組NVLink埠,每一組NVLink埠分別和一個NV-Switch(匯流排通信協定交換模組)相連,共有8×6 NVLink;每個NV-Switch有十六組NVLink埠,除去和內部GPU互連的八組,還有八組用作NVLink光纖通信的擴展,即十六GPU系統。當兩個GPU需要通信時,如GPU1和GPU2,由於每個GPU都有一組NVLink埠與NV-Switch互連,host可通過PEX8725管理六個NV-Switch, 使得GPU1和GPU2的NVLink相通,這樣可以同時有六組NVLink工作來提供GPU之間的資料傳輸,每組NVLink能提供的雙向頻寬約為50GB/s,總頻寬可達300GB/s。 In the topology diagram of the SXM3 GPU system shown in Figure 7, the communication part between the CPU and GPU still uses the structure of three PCIE switches (PEX9797). The first switch is the virtual mode (virtual mode) working mode, and the internal is divided into two Virtual bridges VS0 and VS1, the two virtual bridges respectively connect a set of X16 PCIE to the CPU to achieve the communication bandwidth of 32 PCIE lanes between GPU and CPU; because NVLink optical fiber communication requires a set of X4 PCIEs for NV-Switch management , The second switch is connected to a PEX8749 as a PCIE expansion, and it works in virtual mode as well. One of the virtual bridges VS0 is used to manage and expand NVLink optical fiber communication. A set of X16 PCIE network cards and SSDs are used; the other virtual bridge VS1 For PCIE switch optical fiber communication management, the downstream is connected to three PEX9797 management ports, which are X1 PCIE, and the upstream is connected to the management board through HD mini SAS, which is X4 PCIE. The downstream control port is reserved for PCIE switch optical fiber. The communication expansion is X1 PCIE. The peer-to-peer network communication between GPU and GPU uses NVLink optical fiber communication structure. Each SXM3 GPU has six sets of NVLink ports, and each set of NVLink ports is connected to an NV-Switch (bus communication protocol switch module), a total of 8×6 NVLink; each NV-Switch has sixteen sets of NVLink ports, In addition to the eight groups interconnected with the internal GPU, there are eight groups used as the expansion of NVLink optical fiber communication, that is, the 16 GPU system. When two GPUs need to communicate, such as GPU1 and GPU2, since each GPU has a set of NVLink ports to connect to the NV-Switch, the host can manage six NV-Switches through PEX8725, Make the NVLink of GPU1 and GPU2 communicate, so that there can be six groups of NVLink jobs at the same time to provide data transmission between GPUs. Each group of NVLink can provide a two-way bandwidth of about 50GB/s, and the total bandwidth can reach 300GB/s.

SXM3 GPU系統的拓撲是在SXM2 GPU系統上改進而來,除了原有的優點之外,還具有以下優點: The topology of the SXM3 GPU system is improved from the SXM2 GPU system. In addition to the original advantages, it also has the following advantages:

本發明所述的PCIE交換機的光纖通信模式主要用於多交換機組網,利用管理埠對交換機的動態管理,可以實現多主機(CPU)和多終端(GPU)之間的靈活通信。 The optical fiber communication mode of the PCIE switch of the present invention is mainly used for multi-switch networking, and the dynamic management of the switch by the management port can realize flexible communication between multiple hosts (CPU) and multiple terminals (GPU).

本發明可以解決GPU相互之間的對等網路通信頻寬偏低的問題,可以給GPU和CPU之間提供高頻寬。 The present invention can solve the problem of low frequency bandwidth of peer-to-peer network communication between GPUs, and can provide high frequency bandwidth between GPU and CPU.

本發明將二級PCIE交換機通過網卡接入網路中後,可實現同一網路中的不同GPU系統間的資料直接交換,而不需要經過CPU及其記憶體,極大地提高了GPU系統間的資料交換能力。 After the second-level PCIE switch is connected to the network through the network card, the present invention can realize the direct exchange of data between different GPU systems in the same network without passing through the CPU and its memory, which greatly improves the communication between GPU systems. Data exchange capability.

此外,本發明將GPU間的對等網路通信頻寬最大化和均衡化,理論最大雙向頻寬可達300GB/s,還可以再擴展一個八GPU系統,構成一個十六GPU系統,可實現任意兩GPU的對等網路通信;本發明將PCIE switch(高速串列電腦擴展匯流排標準交換機組150)的光纖通信和NVLink switch的光纖通信管理鏈路合在一個switch(圖中PEX8749,高速串列電腦擴展匯流排標準擴展模組)中,簡化了板子結構。 In addition, the present invention maximizes and equalizes the communication bandwidth of the peer-to-peer network between GPUs. The theoretical maximum two-way bandwidth can reach 300GB/s, and an eight-GPU system can be expanded to form a sixteen-GPU system. The peer-to-peer network communication of any two GPUs; the present invention combines the optical fiber communication of PCIE switch (high-speed serial computer expansion bus standard switch group 150) and the optical fiber communication management link of NVLink switch into one switch (PEX8749 in the figure, high-speed In the serial computer expansion bus standard expansion module), the board structure is simplified.

綜上所述,本發明有效克服了現有技術中的種種缺點而具高度產業利用價值。 In summary, the present invention effectively overcomes various shortcomings in the prior art and has a high industrial value.

上述實施例僅例示性說明本發明的原理及其功效,而非用於限 制本發明。任何熟悉此技術的人士皆可在不違背本發明的精神及範疇下,對上述實施例進行修飾或改變。因此,舉凡所屬技術領域中具有通常知識者在未脫離本發明所揭示的精神與技術思想下所完成的一切等效修飾或改變,仍應由本發明的權利要求所涵蓋。 The above-mentioned embodiments only exemplarily illustrate the principles and effects of the present invention, and are not intended to limit The invention is made. Anyone familiar with this technology can modify or change the above-mentioned embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed in the present invention should still be covered by the claims of the present invention.

100      圖形處理器系統 110      中央處理器 120      圖形處理器 130      匯流排通信協定交換機 140      管理板100 Graphics processor system 110 Central processor 120 Graphics processor 130 Bus communication protocol switch 140 Management board

Claims (9)

一種圖形處理器系統,包含:一中央處理器;複數個圖形處理器,與該中央處理器通信相連;一匯流排通信協定交換機,與該些圖形處理器相連,用於實現各個該圖形處理器之間的相互通信;一管理板,與該匯流排通信協定交換機分別通信相連,用於對該匯流排通信協定交換機進行管理,其中該匯流排通信協定交換機包括:至少一匯流排通信協定交換模組,該匯流排通信協定交換模組包含6個匯流排通信協定交換模塊;各該些匯流排通信協定交換模塊皆設有16組匯流排通信協定埠;各該些匯流排通信協定交換模塊的每個該些組匯流排通信協定埠皆用於連接一圖形處理器或擴展;一匯流排通信協定管理模組,與各該些匯流排通信協定交換模塊相連,用於管理各該些匯流排通信協定交換模塊的各該些組匯流排通信協定埠之間的通信。 A graphics processor system, comprising: a central processing unit; a plurality of graphics processing units, which are connected in communication with the central processing unit; a bus communication protocol switch, which is connected to the graphics processing units and used to implement each of the graphics processing units A management board, which is respectively connected to the bus communication protocol switch and used to manage the bus communication protocol switch, wherein the bus communication protocol switch includes: at least one bus communication protocol switch module Group, the bus communication protocol switching module includes 6 bus communication protocol switching modules; each of the bus communication protocol switching modules is equipped with 16 sets of bus communication protocol ports; each of the bus communication protocol switching modules Each of these sets of bus communication protocol ports is used to connect a graphics processor or an extension; a bus communication protocol management module is connected to each of the bus communication protocol switching modules for managing each of the buses Communication between the groups of bus communication protocol ports of the communication protocol exchange module. 如請求項1所述的圖形處理器系統,該些匯流排通信協定交換模塊中的8組匯流排通信協定埠用於一一對應連接8個該些圖形處理器,各該些匯流排通信協定交換模塊中的剩餘8組該些匯流排通信協定埠用於擴展。 For the graphics processor system according to claim 1, the 8 sets of bus communication protocol ports in the bus communication protocol switching modules are used to connect 8 graphics processors one by one, each of the bus communication protocols The remaining 8 sets of these bus communication protocol ports in the switch module are used for expansion. 如請求項1所述的圖形處理器系統,該圖形處理器系統更包含: 一高速串列電腦擴展匯流排標準交換機組,與該中央處理器和複數個圖形處理器分別相連,用於擴展該中央處理器的一高速串列電腦擴展匯流排標準埠,實現該些圖形處理器與該中央處理器之間的相互通信;一高速串列電腦擴展匯流排標準擴展模組,與該高速串列電腦擴展匯流排標準交換機組、該匯流排通信協定管理模組和該管理板分別通信相連;該管理板通過該高速串列電腦擴展匯流排標準擴展模組對該高速串列電腦擴展匯流排標準交換機組和該匯流排通信協定交換機進行管理。 The graphics processor system according to claim 1, the graphics processor system further comprising: A high-speed serial computer expansion bus standard switch group is respectively connected to the central processing unit and a plurality of graphics processors, and a high-speed serial computer expansion bus standard port used to expand the central processing unit to realize the graphics processing The mutual communication between the processor and the central processor; a high-speed serial computer expansion bus standard expansion module, and the high-speed serial computer expansion bus standard switch group, the bus communication protocol management module and the management board They are respectively connected by communication; the management board manages the high-speed serial computer expansion bus standard switch group and the bus communication protocol switch through the high-speed serial computer expansion bus standard expansion module. 如請求項3所述的圖形處理器系統,其中該高速串列電腦擴展匯流排標準交換機組包括:一第一高速串列電腦擴展匯流排標準交換機,與該中央處理器相連;一第二高速串列電腦擴展匯流排標準交換機,與該第一高速串列電腦擴展匯流排標準交換機相連;該第二高速串列電腦擴展匯流排標準交換機下行連接複數個圖形處理器;一第三高速串列電腦擴展匯流排標準交換機,與該第一高速串列電腦擴展匯流排標準交換機相連;該第三高速串列電腦擴展匯流排標準交換機下行連接複數個圖形處理器;該管理板與該第一高速串列電腦擴展匯流排標準交換機、該第二高速串列電腦擴展匯流排標準交換機和該第三高速串列電腦擴展匯流排標準交換機分別相連,用於對該第一高速串列電腦擴展匯流排標準交換機、該第二高速串列電腦擴展匯流排標準交換機和該第三高速串列電腦擴展匯流排標準交換機進行管理設置。 The graphics processor system according to claim 3, wherein the high-speed serial computer expansion bus standard switch group includes: a first high-speed serial computer expansion bus standard switch connected to the central processing unit; and a second high-speed serial computer expansion bus standard switch The serial computer expansion bus standard switch is connected to the first high-speed serial computer expansion bus standard switch; the second high-speed serial computer expansion bus standard switch is connected downstream to a plurality of graphics processors; a third high-speed serial The computer expansion bus standard switch is connected to the first high-speed serial computer expansion bus standard switch; the third high-speed serial computer expansion bus standard switch is connected downstream to a plurality of graphics processors; the management board is connected to the first high-speed The serial computer expansion bus standard switch, the second high-speed serial computer expansion bus standard switch and the third high-speed serial computer expansion bus standard switch are respectively connected for the first high-speed serial computer expansion bus The standard switch, the second high-speed serial computer expansion bus standard switch and the third high-speed serial computer expansion bus standard switch perform management settings. 如請求項4所述的圖形處理器系統,其中該第一高速串列電腦擴展匯流排標準交換機包含:至少一第一主機埠,用於與該中央處理器通信相連;至少2個第一光纖通信埠,用於分別與該第二高速串列電腦擴展匯流排標準交換機和該第三高速串列電腦擴展匯流排標準交換機通信相連;至少一第一上行埠,用於與該管理板通信相連。 The graphics processor system according to claim 4, wherein the first high-speed serial computer expansion bus standard switch includes: at least one first host port for communicating with the central processing unit; and at least two first optical fibers The communication ports are used to communicate with the second high-speed serial computer expansion bus standard switch and the third high-speed serial computer expansion bus standard switch respectively; at least one first uplink port is used to communicate with the management board . 如請求項5所述的圖形處理器系統,其中該第二高速串列電腦擴展匯流排標準交換機包括:至少一第二光纖通信埠,用於與該第一高速串列電腦擴展匯流排標準交換機的一第一光纖通信埠通信相連;至少一第二上行埠,用於與該管理板通信相連;至少一第二下行埠,用於與該圖形處理器通信相連。 The graphics processor system according to claim 5, wherein the second high-speed serial computer expansion bus standard switch includes: at least one second optical fiber communication port for communicating with the first high-speed serial computer expansion bus standard switch At least one second uplink port is used to communicate with the management board; at least one second downlink port is used to communicate with the graphics processor. 如請求項6所述的圖形處理器系統,其中該第三高速串列電腦擴展匯流排標準交換機包括:至少一第三光纖通信埠,用於與該第一高速串列電腦擴展匯流排標準交換機的另一第一光纖通信埠通信相連;至少一第三上行埠,用於與該管理板通信相連;至少一第三下行埠,用於與該圖形處理器通信相連。 The graphics processor system according to claim 6, wherein the third high-speed serial computer expansion bus standard switch includes: at least one third optical fiber communication port for communicating with the first high-speed serial computer expansion bus standard switch At least one third uplink port is used to communicate with the management board; at least one third downlink port is used to communicate with the graphics processor. 如請求項7所述的圖形處理器系統,其中該第二高速串列電腦擴展匯流排標準交換機的一第二下行埠用於連接一網卡或一固態硬碟;或該第三高速串列電腦擴展匯流排標準交換機的一第三下行埠用於連接一網卡或一固態硬碟。 The graphics processor system according to claim 7, wherein a second downstream port of the second high-speed serial computer expansion bus standard switch is used for connecting a network card or a solid state drive; or the third high-speed serial computer A third downstream port of the extended bus standard switch is used to connect a network card or a solid state drive. 如請求項4所述的圖形處理器系統,其中該第一高速串列電腦擴展匯流排標準交換機、該第二高速串列電腦擴展匯流排標準交換機和該第三高速串列電腦擴展匯流排標準交換機均工作於虛擬模式;該管理板對該第一高速串列電腦擴展匯流排標準交換機、該第二高速串列電腦擴展匯流排標準交換機和該第三高速串列電腦擴展匯流排標準交換機進行動態管理。The graphics processor system according to claim 4, wherein the first high-speed serial computer expansion bus standard switch, the second high-speed serial computer expansion bus standard switch, and the third high-speed serial computer expansion bus standard The switches all work in virtual mode; the management board performs operations on the first high-speed serial computer expansion bus standard switch, the second high-speed serial computer expansion bus standard switch, and the third high-speed serial computer expansion bus standard switch Dynamic management.
TW107143644A 2018-12-05 2018-12-05 Graphics processor system TWI698833B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
TW107143644A TWI698833B (en) 2018-12-05 2018-12-05 Graphics processor system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
TW107143644A TWI698833B (en) 2018-12-05 2018-12-05 Graphics processor system

Publications (2)

Publication Number Publication Date
TW202022793A TW202022793A (en) 2020-06-16
TWI698833B true TWI698833B (en) 2020-07-11

Family

ID=72175668

Family Applications (1)

Application Number Title Priority Date Filing Date
TW107143644A TWI698833B (en) 2018-12-05 2018-12-05 Graphics processor system

Country Status (1)

Country Link
TW (1) TWI698833B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120173755A1 (en) * 2005-11-30 2012-07-05 Microsoft Corporation Multi-User Display Proxy Server
TW201314703A (en) * 2011-06-14 2013-04-01 Nvidia Corp A clamped bit line read circuit
CN206039604U (en) * 2014-12-11 2017-03-22 埃科亚特姆公司 System for be used for purchaseing electronic equipment from consumer
CN107450987A (en) * 2017-07-28 2017-12-08 山东超越数控电子有限公司 A kind of heterogeneous server of High Availabitity

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120173755A1 (en) * 2005-11-30 2012-07-05 Microsoft Corporation Multi-User Display Proxy Server
TW201314703A (en) * 2011-06-14 2013-04-01 Nvidia Corp A clamped bit line read circuit
CN206039604U (en) * 2014-12-11 2017-03-22 埃科亚特姆公司 System for be used for purchaseing electronic equipment from consumer
CN107450987A (en) * 2017-07-28 2017-12-08 山东超越数控电子有限公司 A kind of heterogeneous server of High Availabitity

Also Published As

Publication number Publication date
TW202022793A (en) 2020-06-16

Similar Documents

Publication Publication Date Title
CN109408451B (en) Graphic processor system
CN100461140C (en) Method and system for supporting multiple graphic processing units
CN100481050C (en) Method and system for supporting multiple graphic processing units
WO2024221975A1 (en) Converged infrastructure system, non-volatile memory system, and memory resource acquisition method
CN108845970A (en) A kind of device and method of free switching GPU topology server
CN109410117B (en) Graphics processor system
CN101937326A (en) Multi-CPU parallel video processing system, cascade system and method thereof
CN107430574A (en) Method and apparatus for the optimization of the IO of analysis system, processing and bandwidth of memory
CN204650513U (en) Distributed structure/architecture equipment and serial port circuit thereof
WO2018213232A1 (en) Reconfigurable server and server rack with same
CN110221999A (en) It is realized based on standard PCIE and supports GPU and M2 SSD extension adapter and implementation method
CN107748726A (en) A kind of GPU casees
CN206249150U (en) A kind of storage server
CN117851283A (en) A distributed memory orthogonal architecture based on CXL
CN209248436U (en) An expansion board and server
TWI698833B (en) Graphics processor system
CN119718691B (en) Bandwidth allocation method, module, server, electronic device, medium and product
CN101937410A (en) Control apparatus and method thereof
CN209248518U (en) A solid-state hard disk expansion board and server
CN103457880A (en) Switch system and method of operating a switch
CN101290556A (en) Disk Array Shared Device
CN106528475A (en) Multi-channel computer system supporting single octal and dual quad dynamic partition
TWI690789B (en) Graphic processor system
CN210955055U (en) Display control computer motherboard framework based on soar
WO2024139593A1 (en) Computer device and computer system