AI module supporting cooperative work of multiple AI processors under mercuration platform and working method

Abstract

The invention relates to an AI module supporting cooperative work of multiple AI processors under a mercuric chloride platform and a working method. The AI module comprises a first mercuric chloride AI processor, a second mercuric chloride AI processor, a BOOT starting circuit, a PCIE switching circuit, a BMC health management circuit, a power conversion circuit and a peripheral interface circuit. Wherein the BOOT starting circuit is respectively connected with the first mercuric chloride AI processor and the second mercuric chloride AI processor and is used for starting the first mercuric chloride AI processor and the second mercuric chloride AI processor, and the BMC health management circuit is respectively connected with the first mercuric chloride AI processor and the second mercuric chloride AI processor and is used for collecting temperature and voltage; the PCIE switching circuit is respectively connected with the first mercuric chloride AI processor and the second mercuric chloride AI processor and is used for converting two paths of PCIE 3.0 * 4 into PCIE 3.0 * 8 to communicate with a mainboard, and the peripheral interface circuit is respectively connected with the first mercuric chloride AI processor and the second mercuric chloride AI processor and is used for connecting various peripherals. According to the invention, the onboard AI processor can be expanded through the PCIE bridge chip, so that the computing power of the module is improved.

Description

Technical Field

[0001] The present invention belongs to the field of computer processing technology, and in particular to an AI module and working method supporting collaborative work of multiple AI processors under an Ascend platform. Background Art

[0002] With the rapid development of artificial intelligence (AI), AI technology has been widely used in various fields, such as intelligent transportation, smart cities, smart retail, face recognition, video analysis, etc. AI technology has greatly improved the performance of image recognition, target detection and natural language processing. Compared with related foreign companies, although domestic AI processors started late, many companies have been vigorously developing self-developed AI processors.

[0003] With the rise of related technologies such as smart driving and smart drones, artificial intelligence computing devices equipped with intelligent target detection and speech recognition algorithms have gradually entered a stage of rapid development. However, most of the current artificial intelligence devices have fixed computing power and fixed working modes. They are unable to adjust the computing power and working mode according to the usage scenarios, which easily leads to a waste of computing resources. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and propose an AI module and working method under the Ascend platform that supports the collaborative work of multiple AI processors. The onboard AI processor can be expanded through a PCIE bridge chip to improve the module computing power; the RC mode (Root Complex, PCIE master) and EP mode (EndPoint, PCIE slave) can be switched according to different needs. It can be used in conjunction with the motherboard to form a reinforced intelligent computing computer, and can also run the operating system independently.

[0005] The present invention solves the technical problem by adopting the following technical solutions:

[0006] The AI ​​module under the Ascend platform that supports the collaborative work of multiple AI processors includes the first Ascend AI processor, the second Ascend AI processor, the BOOT startup circuit, the PCIE switching circuit, the BMC health management circuit, the power conversion circuit and the peripheral interface circuit; wherein the BOOT startup circuit is respectively connected to the first Ascend AI processor and the second Ascend AI processor for starting the first Ascend AI processor and the second Ascend AI processor, the BMC health management circuit is respectively connected to the first Ascend AI processor and the second Ascend AI processor for collecting temperature and voltage; the PCIE switching circuit is respectively connected to the first Ascend AI processor and the second Ascend AI processor for converting two PCIE 3.0×4 into PCIE 3.0×8 to communicate with the motherboard, the peripheral interface circuit is respectively connected to the first Ascend AI processor and the second Ascend AI processor for connecting various peripherals, and the power conversion circuit is used to convert 12V power into power of other sizes.

[0007] Moreover, the circuit board is a 6U CPEX structure, and the circuit board power supply is input through a CPEX interface.

[0008] Moreover, the CPEX interface is connected to the PCIE switching circuit through PCIE 3.0×8, the CPEX interface is connected to the power conversion circuit through the +12V power interface, the CPEX interface is connected to the peripheral interface circuit, and the CPEX interface is connected to the BMC health management circuit through I2C.

[0009] Moreover, the power conversion circuit includes a surge protector, an SM4644 power module and a 5430 power module, wherein the input end of the power conversion circuit is connected to the CPEX interface, and the surge protector is used to connect the cooling fan; the SM4644 power module is used to provide power for the PCIE bridge chip, the SD card, the clock BUFFER, the microcontroller and the PHY chip in the first Ascend AI processor and the second Ascend AI processor; the 5430 power module is used for USB power supply and HDMI power supply of the peripheral interface circuit.

[0010] Moreover, the PCIE switching circuit is connected to the upstream mainboard device through the PORT 0 port of SM8724; the PCIE switching circuit is connected to the first Ascend AI processor and the second Ascend AI processor through the PORT 1 port and the PORT 2 port respectively.

[0011] Moreover, the BMC health management module includes a temperature sensor, a voltage divider circuit and a control chip, wherein the control chip is respectively connected to the temperature sensor and the voltage divider circuit, the temperature sensor is used to collect the module core chip temperature; the voltage divider circuit communicates with the mainboard using the I2C interface through the collected power supply voltage, and the control chip uses the single-chip microcomputer GD32F103CBT6 to interact with the mainboard through the I2C isolation chip NSI8100N.

[0012] Furthermore, the circuit board further comprises two RGMII controllers for data transmission between the Ethernet MAC layer and the PHY layer, wherein the MDIO interface is used for configuring and managing the PHY chip of the external Ethernet interface.

[0013] A working method of an AI module supporting collaborative work of multiple AI processors under an Ascend platform includes the following steps:

[0014] Step 1: Power on the module, and the power conversion circuit converts the 12V main power supply into the working level of each circuit;

[0015] Step 2: The module determines the working mode. If the module works in the EP mode, proceed to step 3. If the module works in the EP mode, proceed to step 4.

[0016] Step 3: The PCIE switching circuit works. The PCIE bridge chip converts two PCIE 3.0×4 into PCIE 3.0×8 to communicate with the motherboard. The motherboard recognizes the PCIE device and works.

[0017] Step 4: The BOOT startup circuit works, and the module is started through the SD card. The module obtains the module working information or remotely controls the module through the peripheral interface circuit.

[0018] The advantages and positive effects of the present invention are:

[0019] 1. The present invention includes a first Ascend AI processor, a second Ascend AI processor, a BOOT startup circuit, a PCIE switching circuit, a BMC health management circuit, a power conversion circuit and a peripheral interface circuit; wherein the BOOT startup circuit is respectively connected to the first Ascend AI processor and the second Ascend AI processor for starting the first Ascend AI processor and the second Ascend AI processor, the BMC health management circuit is respectively connected to the first Ascend AI processor and the second Ascend AI processor for collecting temperature and voltage; the PCIE switching circuit is respectively connected to the first Ascend AI processor and the second Ascend AI processor for converting two PCIE 3.0×4 into PCIE 3.0×8 to communicate with the mainboard, the peripheral interface circuit is respectively connected to the first Ascend AI processor and the second Ascend AI processor for connecting various peripherals, and the power conversion circuit is used to convert 12V power into power of other sizes. The present invention can improve the module computing power by expanding the onboard AI processor through the PCIE bridge chip; the RC mode (Root Complex, PCIE master) and the EP mode (EndPoint, PCIE slave) can be switched according to different needs, and it can be used in conjunction with the motherboard to form a reinforced intelligent computing computer, and can also run the operating system independently.

[0020] 2. The present invention can realize the collaborative work of multiple AI processors, and can achieve higher computing performance under the same conditions through chip expansion and resource allocation.

[0021] 3. The present invention can realize the switching between RC mode and EP mode to adapt to different application scenarios. The working mode is configured by a dip switch, and the configuration method is fast and convenient.

[0022] 4. In the RC mode, the present invention saves the operating system and files in the SD card, can independently run the operating system and artificial intelligence applications, and can also be used as a slave of the host system, connected to the main board through a network interface, executes the instructions issued by the host and returns the result information, thereby reducing the working pressure of the host and having high flexibility.

[0023] 5. In EP mode, the present invention can cooperate with domestic high-performance CPUs to form intelligent computers with higher computing power, thereby realizing AI empowerment of existing reinforced computers.

[0024] 6. The present invention can quickly transplant and deploy projects such as image recognition, natural language processing, video encoding and decoding running on other platforms to this platform through the developed software tool chain.

[0025] 7. The present invention can realize real-time monitoring of the operating temperature and voltage of the AI ​​chip, and alarm when an abnormality occurs. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a structural diagram of the present invention;

[0027] Figure 2 This is a power supply circuit diagram of the present invention;

[0028] Figure 3 This is a PCIE switching circuit diagram of the present invention;

[0029] Figure 4 It is the BOOT startup circuit diagram of the present invention;

[0030] Figure 5 This is a principle block diagram of the network signal link of the present invention;

[0031] Figure 6 Developing an environment deployment process for the present invention;

[0032] Figure 7 The driver and firmware installation process of the present invention;

[0033] Figure 8 This is the BMC health management circuit diagram of the present invention;

[0034] Fig. 9 It is a schematic diagram of the working environment of the RC mode of the present invention.

[0035] Fig.10 It is a schematic diagram of the working environment of the EP mode of the present invention. DETAILED DESCRIPTION

[0036] The present invention is further described in detail below with reference to the accompanying drawings.

[0037] The construction idea of ​​the present invention is: to realize the collaborative work of multiple AI processors, and to achieve higher computing performance under the same conditions through chip expansion and resource allocation; to realize switching between RC mode and EP mode to adapt to different application scenarios, and the working mode is configured through the dip switch, and the configuration method is fast and convenient; in RC mode, it can independently run the operating system and run artificial intelligence applications, and can also be used as a slave of the host system, with high flexibility; in EP mode, it can cooperate with domestic high-performance CPUs to form an intelligent computing computer with high computing power, and realize AI empowerment of existing reinforced computers; through the developed software tool chain, image recognition, natural language processing, video encoding and decoding and other projects running on other platforms can be quickly transplanted and deployed to this platform; real-time monitoring of the operating temperature and voltage of the AI ​​chip can be realized, and an alarm will be issued when an abnormality occurs.

[0038] AI modules under the Ascend platform that support the collaboration of multiple AI processors, such as Figure 1 As shown, it includes a first Ascend AI processor, a second Ascend AI processor, a BOOT startup circuit, a PCIE switching circuit, a BMC health management circuit, a power conversion circuit and a peripheral interface circuit; wherein the BOOT startup circuit is respectively connected to the first Ascend AI processor and the second Ascend AI processor for starting the first Ascend AI processor and the second Ascend AI processor, and the BMC health management circuit is respectively connected to the first Ascend AI processor and the second Ascend AI processor for collecting temperature and voltage; the PCIE switching circuit is respectively connected to the first Ascend AI processor and the second Ascend AI processor for converting two PCIE 3.0×4 into PCIE 3.0×8 to communicate with the motherboard, the peripheral interface circuit is respectively connected to the first Ascend AI processor and the second Ascend AI processor for connecting various peripherals, and the power conversion circuit is used to convert 12V power into power of other sizes.

[0039] like Figure 1 As shown, the module is a 6U CPEX structure. The dimensions of the CPEX 6U board are 233.35mm high, 160.02mm wide, and 1.57mm thick. The circuit board power is input through the CPEX interface; the CPEX interface is connected to the PCIE switching circuit through PCIE 3.0×8, the CPEX interface is connected to the power conversion circuit through the +12V power interface, the CPEX interface is connected to the peripheral interface circuit, and the CPEX interface is connected to the BMC health management circuit through I2C.

[0040] The AI ​​processor of the present invention is Huawei Ascend 310B, the PCIE bridge chip is SM8724 of Shenzhen Guowei Electronics Co., Ltd., the PHY chip is YT8521SH of Yutai Microelectronics Co., Ltd., and the single-chip microcomputer model is GigaDevice's GD32F103CBT6; the CPU is Feiteng D2000, and the operating system is Kylin V10; the peripheral interfaces include Gigabit network port, USB interface, HDMI interface, serial port and SD card slot.

[0041] like Figure 2 As shown, the power conversion circuit includes a surge protector, an SM4644 power module and a 5430 power module, wherein the input end of the power conversion circuit is connected to the CPEX interface, and the surge protector is used to connect the cooling fan; the SM4644 power module is used to provide power for the PCIE bridge chip, the SD card, the clock BUFFER, the microcontroller and the PHY chip in the first Ascend AI processor and the second Ascend AI processor; the 5430 power module is used for USB power supply and HDMI power supply of the peripheral interface circuit.

[0042] The power conversion circuit can provide 4 power outputs, with a maximum current of 4A per channel, and supports parallel output, providing 0.9V and 1.8V power supply support for the PCIE bridge chip, of which 0.9V voltage is a dual-channel parallel output; and provides 3.3V voltage for the SD card, clock BUFFER, microcontroller and PHY chip in the first Ascend AI processor and the second Ascend AI processor. The module has a cooling fan power supply interface reserved for two external 12V fans. The 5430 power chip provides 5V power for the peripheral interface circuits USB and HDMI.

[0043] The PCIE bridge uses the SM8724 PCIE switching chip from Shenzhen Guowei Electronics Co., Ltd., which has 24 channels and up to 6 ports. It can realize data exchange of PCIE3.0 protocol and is backward compatible.

[0044] like Figure 3 As shown, the mainboard device is the upstream device, connected through PORT 0 of SM8724, occupying 8 channels; the downstream device is 310B, each occupying 4 channels, so a maximum of 4 310B modules can be connected. This design only uses 2, occupying PORT 1 and PORT 2, and the other two ports can be used for subsequent expansion.

[0045] according to Figure 3Assignment, configure the SM8724 related pins, configure STRAP_STN0_PORTCFG[1:0] to Z0, configure STRAP_STN0_PORTCFG[1:0] to Z0, configure STRAP_UPSTRM_PORTSEL[2:0] to 000, and you can realize PORT 0 as the x8 channel upstream port and PORT 1 / 2 / 9 / 10 as the x4 channel downstream port.

[0046] like Figure 4 As shown, the BMC health management module includes a temperature sensor, a voltage divider circuit and a control chip, wherein the control chip is connected to the temperature sensor and the voltage divider circuit respectively, and the temperature sensor is used to collect the module core chip temperature; the voltage divider circuit uses the collected power supply voltage to communicate with the mainboard using the I2C interface, and the control chip uses the single-chip microcomputer GD32F103CBT6 to interact with the mainboard through the I2C isolation chip NSI8100N.

[0047] Part of the health information comes from the module core chip temperature collected by the temperature sensor; the other part comes from the power supply voltage collected by the voltage divider circuit, and uses the I2C interface to communicate with the mainboard. The control chip uses GigaDevice's single-chip microcomputer GD32F103CBT6, which interacts with the mainboard through the I2C isolation chip NSI8100N. In addition, a serial port is reserved as a backup communication method.

[0048] like Figure 5 As shown, the switching between the module RC mode and EP mode can be configured through the PCIE_EP_RC_FLAG pin, and the boot medium can be selected by setting the BOOT_SEL[2:0] pin level on the module.

[0049] When PCIE_EP_RC_FLAG is high, the module works in EP mode, which only supports PCIe boot mode, that is, the value of BOOT_SEL[2:0] must be 010; when PCIE_EP_RC_FLAG is low, the module works in RC mode, which supports EMMC, SATA, and PCIe boot modes, that is, the value of BOOT_SEL[2:0] must be 101. The switching of BOOT mode is achieved through a 4-bit DIP switch. When the switch is closed, the value of this pin is 0, and when the switch is open, the value of this pin is 1. The BOOT switch part is designed at the front end of the module, and there is no need to remove the module from the chassis when switching the boot mode.

[0050] The module integrates two RGMII controllers for data transmission between the Ethernet MAC layer and the PHY layer. It also has an MDIO interface to configure and manage the PHY chip of the external Ethernet interface. The software can read and write the internal registers of the PHY chip through the MDIO interface to achieve control and management of the PHY chip, such as reset control, self-loop control, speed selection, etc.

[0051] The PHY chip uses the YT8521SH single-port Gigabit Ethernet physical layer transceiver from Yutai Microelectronics. The signal link principle block diagram between 310B and the PHY chip is as follows: Figure 6 As shown, 310B accesses and reads the internal registers of the PHY chip through the MDIO interface, and exchanges data with the PHY chip through the RGMII interface. During this period, it is necessary to control the level of the RGMII_RXD[3:1] pin to control the CFG_MODE configuration. Setting RGMII_RXD[3:1] to 000 indicates that it works in the UTP<->RGMII mode. In addition, the RGMII voltage selection of the PHY chip needs to be configured by controlling the level of the CFG_LDO[1:0] pin. Setting CFG_LDO[1:0] to 10 indicates that the built-in 1.8V power supply of the PHY chip is used. The data signal is output through the TRX P / N differential signal pair of the PHY chip, and a TVS diode needs to be added for electrostatic protection, and then output to the external connector after passing through the network transformer.

[0052] A working method of an AI module supporting collaborative work of multiple AI processors under an Ascend platform includes the following steps:

[0053] Step 1: Power on the module, and the power conversion circuit converts the 12V main power supply into the working level of each circuit.

[0054] Step 2: The module determines the working mode. If the module works in the EP mode, proceed to step 3; if the module works in the EP mode, proceed to step 4.

[0055] When the module works in EP mode, it can carry two Ascend 310B AI processors onboard, and convert two PCIE 3.0×4 to PCIE 3.0×8 through the PCIE bridge chip to communicate with the motherboard, and it is backward compatible. Each Ascend 310B is equipped with a peripheral interface, and a switch is set to switch between RC mode and EP mode. In EP mode, the two modules act as PCIE slave devices of the motherboard for the motherboard CPU to call; in RC mode, 310B has a built-in CPU core and can run the operating system independently. It is equipped with peripheral interfaces including Gigabit network port, USB port, HDMI port and serial port. The operating system and files are saved in the SD card, which is convenient for system environment transplantation and file management, so that it can run artificial intelligence applications independently without the motherboard. In RC mode, the motherboard can obtain acceleration module information through the serial port or remotely control the acceleration module through the network port.

[0056] Step 3: The PCIE switching circuit works. The PCIE bridge chip converts two PCIE 3.0×4 into PCIE 3.0×8 to communicate with the motherboard. The motherboard recognizes the PCIE device and starts working.

[0057] like Fig.10 As shown, if the module works in EP mode, it needs to be used in conjunction with an external host with a CPU. At this time, the module's peripheral interfaces are all invalid, and the module only communicates with the host through the PCIE bus. For the target detection program, the test program is run in the host system. The program first reads the video stream collected by the camera in real time. If the module is used for hardware decoding, the video stream data is sent to the module through the PCIE bus for decoding processing. After the decoded data is resized, the audit network calculation is performed, and the calculation result is sent back to the host through the PCIE bus. After post-processing by the host, the final result can be output through the host terminal; if the CPU is used for decoding, the decoded data is pre-processed by the CPU and sent to the module through the PCIE bus for neural network calculation. The calculation result is then sent back to the host for post-processing. The final result can be observed in real time through the display connected to the host or output through the terminal.

[0058] Step 4: The BOOT startup circuit works, and the module is started through the SD card. The module obtains the module working information or remotely controls the module through the peripheral interface circuit.

[0059] Before using the EP mode, you need to deploy a development environment. Based on the heterogeneous computing architecture CANN, the module can support a variety of AI frameworks, including MindSpore, PyTorch, TensorFlow, etc. Through the developed software tool chain, projects such as image recognition, natural language processing, and video encoding and decoding can be quickly transplanted and deployed to the platform.

[0060] The process of building the CANN development and operation environment on the module is as follows Figure 7 As shown, first you need to check the environment, check whether the NPU chip is correctly read by the system according to the system display information, and prepare the driver package and firmware package that match the operating system and CPU. Then create an installation user, install the board driver and firmware, and the driver and firmware installation process is as follows Figure 8 As shown in the figure, after the driver is successfully installed, the npu-smiinfo command will display the echo information, which includes the driver version, the NPU chip in place, and the video memory usage rate. After that, you need to install the CANN_Toolkit development kit package, which mainly includes the tool chain and library files for model development and reasoning using the Ascend chip. Before that, you need to install the Python environment and related dependent libraries. Finally, you can selectively install the required deep learning framework and related plug-ins according to the application situation.

[0061] like Fig. 9 As shown in the figure, if the module works in RC mode, the module will automatically start up after the whole machine is powered on. When starting up, the module reads the operating system and file system from the SD card, and the peripheral interface is activated. The display can be connected through the HDMI interface, the mouse and keyboard can be connected through the USB interface, and the network interface can be connected to the local area network. The host can control the module through the internal serial port of the chassis or through the network remote terminal after networking. For the target detection program, the module first reads the video stream collected by the camera in real time, and then calculates and processes the data according to the model reasoning process. The final detection result can be observed in real time through the display directly connected to the module or output through the remote terminal of the host. BMC health management information can send request instructions through the host terminal, communicate with the microcontroller through the I2C bus or serial port, and the microcontroller returns the current status information.

[0062] Step 5. After the power-on configuration is completed, the BMC health management circuit works, providing an I2C interface to communicate with the mainboard, and can detect the module core chip temperature and power supply voltage, and alarm when abnormal.

[0063] It should be emphasized that the embodiments described in the present invention are illustrative rather than restrictive. Therefore, the present invention includes but is not limited to the embodiments described in the specific implementation manner. Any other implementation manners derived by those skilled in the art based on the technical solution of the present invention also fall within the scope of protection of the present invention.

Claims

1. The AI ​​module on the Ascend platform supports the collaborative work of multiple AI processors, which features: It includes a first Ascend AI processor, a second Ascend AI processor, a BOOT startup circuit, a PCIE switching circuit, a BMC health management circuit, a power conversion circuit and a peripheral interface circuit; wherein the BOOT startup circuit is connected to the first Ascend AI processor and the second Ascend AI processor respectively, and is used to start the first Ascend AI processor and the second Ascend AI processor; the BMC health management circuit is connected to the first Ascend AI processor and the second Ascend AI processor respectively, and is used to collect temperature and voltage; the PCIE switching circuit is connected to the first Ascend AI processor and the second Ascend AI processor respectively, and is used to convert two PCIE 3.0×4 into PCIE 3.0×8 to communicate with the motherboard; the peripheral interface circuit is connected to the first Ascend AI processor and the second Ascend AI processor respectively, and is used to connect various peripherals; the power conversion circuit is used to convert 12V power into power of other sizes.

2. The AI ​​module supporting collaborative work of multiple AI processors under the Ascend platform according to claim 1, characterized in that: The circuit board is a 6U CPEX structure, and the circuit board power supply is input through a CPEX interface.

3. The AI ​​module supporting collaborative work of multiple AI processors under the Ascend platform according to claim 2, characterized in that: The CPEX interface is connected to the PCIE switching circuit through PCIE 3.0×8, the CPEX interface is connected to the power conversion circuit through the +12V power interface, the CPEX interface is connected to the peripheral interface circuit, and the CPEX interface is connected to the BMC health management circuit through I2C.

4. The AI ​​module supporting collaborative work of multiple AI processors under the Ascend platform according to claim 1, characterized in that: The power conversion circuit includes a surge protector, an SM4644 power module and a 5430 power module, wherein the input end of the power conversion circuit is connected to the CPEX interface, and the surge protector is used to connect to the cooling fan; the SM4644 power module is used to provide power for the PCIE bridge chip, the SD card, the clock BUFFER, the microcontroller and the PHY chip in the first Ascend AI processor and the second Ascend AI processor; the 5430 power module is used for USB power supply and HDMI power supply of the peripheral interface circuit.

5. The AI ​​module supporting collaborative work of multiple AI processors under the Ascend platform according to claim 1, characterized in that: The PCIE switching circuit is connected to the upstream mainboard device through the PORT 0 port of SM8724; the PCIE switching circuit is connected to the first Ascend AI processor and the second Ascend AI processor through the PORT 1 port and the PORT 2 port respectively.

6. The AI ​​module supporting collaborative work of multiple AI processors under the Ascend platform according to claim 1, characterized in that: The BMC health management module includes a temperature sensor, a voltage divider circuit and a control chip, wherein the control chip is connected to the temperature sensor and the voltage divider circuit respectively, and the temperature sensor is used to collect the obtained module core chip temperature; The voltage divider circuit uses the collected power supply voltage to communicate with the mainboard using the I2C interface. The control chip uses the single-chip microcomputer GD32F103CBT6 and interacts with the mainboard through the I2C isolation chip NSI8100N.

7. The AI ​​module supporting collaborative work of multiple AI processors under the Ascend platform according to claim 1, characterized in that: The circuit board also includes two RGMII controllers for data transmission between the Ethernet MAC layer and the PHY layer, wherein the MDIO interface is used to configure and manage the PHY chip of the external Ethernet interface.

8. A working method of an AI module supporting collaborative work of multiple AI processors under the Ascend platform according to any one of claims 1 to 7, characterized in that: The following steps are involved: Step 1: Power on the module, and the power conversion circuit converts the 12V main power supply into the working level of each circuit; Step 2: The module determines the working mode. If the module works in the EP mode, proceed to step 3. If the module works in the EP mode, proceed to step 4. Step 3: The PCIE switching circuit works. The PCIE bridge chip converts two PCIE 3.0×4 into PCIE 3.0×8 to communicate with the motherboard. The motherboard recognizes the PCIE device and works. Step 4: The BOOT startup circuit works, and the module is started through the SD card. The module obtains the module working information or remotely controls the module through the peripheral interface circuit.

Images