A high-computing power processing device

Abstract

The utility model relates to communication technical field discloses a kind of high arithmetic processing equipment, including three box bodies, three the box body is respectively encapsulated with general computing unit, general service unit and AI intelligent computing unit, the back panel of the box body is provided with LRM connector and external interface, the front panel of the box body is equipped with swing handle, the front panel of the box body is provided with debugging maintenance interface, power switch, secret switch, the secret switch is full-sealing self-resetting button switch, the general computing unit includes general computing unit carrier plate, storage unit, management module and GPU module.In the utility model, the LRM modularization design of three units of general computing, general service and AI intelligent computing, through the hardware decoupling and pooling of heterogeneous arithmetic unit, on-demand dynamic combination and task accurate scheduling are realized, the arithmetic type solidification pattern is broken, and the resource utilization is significantly improved.

Description

Technical Field

[0001] This utility model relates to the field of communication technology, and in particular to a high computing power processing device. Background Technology

[0002] High-performance computing devices are high-performance, integrated computing platforms designed specifically for processing massive amounts of data and completing extremely complex computational tasks. Modern artificial intelligence training, scientific simulation, big data analysis, and other tasks require processing data with trillions of parameters or petabytes of data, which traditional computing architectures cannot complete within a reasonable timeframe. Therefore, high-performance computing devices are needed.

[0003] A search revealed Chinese patent publication number CN118171709A, which discloses a high-reliability intelligent computing acceleration module, comprising an Ascend AI processor, LPDDR4X memory, SPI Flash, power conversion circuitry, and a button connector. The Ascend AI processor can operate in both RC and EP modes. The connector used in the module features solderlessness, high signal transmission rate, and low profile, with a total of 328 pins and eight locking screw holes and guide posts, achieving high-precision and high-reliability connections under high vibration conditions while allowing for flexible disassembly. The intelligent computing acceleration module proposed in this invention possesses excellent scalability, allowing for rapid expansion of various communication interfaces via a carrier board as needed, providing a solution for accelerating intelligent algorithm execution in embedded scenarios and meeting the intelligent algorithm processing requirements of various embedded devices.

[0004] However, in actual use, the above-mentioned devices have the following problems: key components cannot be maintained independently, and the computing power type of the fixed architecture equipment is fixed, making it impossible to dynamically configure heterogeneous computing units, resulting in insufficient resource utilization. Utility Model Content

[0005] To overcome the above shortcomings, this utility model provides a high-performance computing device, which aims to improve the problem of insufficient resource utilization caused by the inability to independently maintain key components and the fixed computing power type of fixed architecture devices, which cannot dynamically configure heterogeneous computing units.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a high-performance computing device, comprising three boxes, each of which encapsulates a general computing unit, a general service unit, and an AI intelligent computing unit. The rear panel of each box is provided with an LRM connector and an external interface, and the front panel of each box is equipped with a movable handle. The front panel of each box is also provided with a debugging and maintenance interface, a power switch, and a password destruction switch, wherein the password destruction switch is a fully sealed self-resetting button switch.

[0007] The above technical solution encapsulates the three units in three separate boxes and, through the LRM architecture design, enables them to connect individually to the backplane, achieving a modular effect. This allows for hot-swapping, improved fault recovery speed, and enhanced redundancy, security, and reliability of the equipment.

[0008] As a further description of the above technical solution:

[0009] The general-purpose computing unit includes a general-purpose computing unit carrier board, a storage unit, a management module, and a GPU module. The general-purpose computing unit carrier board integrates an LRM connector, a CPU, memory, a bridge chip, a network controller, a CAN transceiver, and a serial transceiver. The storage unit includes a system disk and a data disk.

[0010] Through the above technical solutions, the general-purpose computing unit achieves a high degree of integration and collaborative operation of computing, storage, and I / O functions through the interconnection design of the processor and bridge chip on the carrier board and the high-speed bus interconnection, which significantly improves data throughput efficiency and resource utilization. Its modular architecture, combined with the fine-grained monitoring of the management module, ensures operational reliability and supports hot-swappable maintenance. The integration of multiple interfaces and GPU expansion capabilities effectively enhance environmental adaptability and parallel processing performance, achieving a unity of high performance, high reliability, and high flexibility while reducing system complexity.

[0011] As a further description of the above technical solution:

[0012] The general service unit includes a general service unit carrier board, a storage unit, a management module, and a GPU module. The general service unit carrier board integrates an LRM connector, a KP920 module, a BMC module, a network controller, a CAN transceiver, and a serial transceiver. The storage unit includes a system disk and a data disk. The general service unit carrier board includes a network unit design and a bus design.

[0013] Through the above technical solutions, the general service unit provides powerful computing power and ultra-high network throughput performance through the KP920 module and high-speed network unit design. It achieves hardware-level intelligent management and energy efficiency optimization by relying on the BMC module and fine bus design. Combined with multi-interface and modular storage expansion capabilities, it significantly improves the efficiency, reliability and environmental adaptability of large-scale service processing.

[0014] As a further description of the above technical solution:

[0015] The AI ​​intelligent computing unit includes an AI intelligent computing unit carrier board and a management module. The AI ​​intelligent computing unit carrier board integrates an LRM connector, two 310P modules, a BMC module, an FPGA module, a serial transceiver, a network transceiver, and a CAN transceiver. The AI ​​intelligent computing unit outputs six LiDARs, three SDI cameras, six video signals, two RS232 interfaces, one RS422 interface, nine CAN interfaces, and two USB interfaces.

[0016] Through the above technical solution, the AI ​​intelligent computing unit integrates two Ascend 310P modules, FPGA video decoding link, BMC management, and high-speed backplane interface into one unit via LRM carrier board, achieving 140TOPs of AI computing power and parallel processing of 128 channels of 1080P@30fps encoding and decoding. FPGA pre-decoding reduces CPU load, and the dual 310P direct connection to PCIe×8 bandwidth is non-blocking. Combined with BMC and management module, it realizes fine power consumption control and fault diagnosis, significantly improving the real-time performance, energy efficiency ratio and environmental adaptability of edge AI applications. At the same time, BMC monitors temperature / power consumption in real time and hot-swappable replacement. Overall, it provides highly reliable and scalable domestic AI acceleration capabilities under miniaturization and low power consumption conditions.

[0017] As a further description of the above technical solution:

[0018] The LRM connector has built-in 10 Gigabit Ethernet interface, Gigabit Ethernet interface, CAN interface, and TSN management interface for high-speed data transmission and management between units.

[0019] Through the above technical solutions, the LRM connector achieves high-speed data exchange between units by integrating 10 Gigabit / Gigabit Ethernet interfaces, and achieves precise device control and time-sensitive network scheduling by relying on CAN interfaces and TSN management interfaces. It integrates service transmission and management functions on a single interface, significantly improving the internal communication efficiency, real-time performance and reliability of the system, while providing a standardized, high-bandwidth and low-latency interconnection foundation for modular devices.

[0020] As a further description of the above technical solution:

[0021] The general computing unit, general service unit, and AI intelligent computing unit are all electrically connected to the password destruction switch.

[0022] Through the above technical solution, the independently connected decryption switches of each unit can significantly improve the physical security and emergency response capability of sensitive information processing, meet the needs of rapid and thorough leakage prevention in high-security scenarios, and provide proactive security protection for domestically produced high-computing-power equipment.

[0023] As a further description of the above technical solution:

[0024] The general computing unit, general service unit, and AI intelligent computing unit are all electrically connected to the power switch, and each of the general computing unit, general service unit, and AI intelligent computing unit is equipped with an aviation connector.

[0025] Through the above technical solution, the independent power switches of each unit, combined with the high-reliability interface design of aviation connectors, can realize precise power supply control and quick plug-and-play maintenance at the module level, significantly improving the flexibility and operation and maintenance efficiency of system power supply management, while ensuring the stability of electrical connections in harsh environments such as vibration and high and low temperatures, providing safety and reliability for high computing power equipment.

[0026] As a further description of the above technical solution:

[0027] The general computing unit, general service unit, and AI intelligent computing unit are each equipped with an independent status indicator light to indicate the power supply status and operating status of their respective units.

[0028] Through the above technical solution, the indicator lights provide real-time feedback on power supply and operating status in a visual manner, which significantly improves the efficiency of module-level fault location and the convenience of system maintenance. It can confirm the status of multiple modules in parallel operation and hot-swap operations, effectively enhancing the maintainability, operational safety and maintenance reliability of high-computing power equipment clusters.

[0029] This utility model has the following beneficial effects:

[0030] 1. In this utility model, the LRM modular design of the three major units of general computing, general service and AI intelligent computing can realize hot-swappable replacement. Through hardware decoupling and pooling of heterogeneous computing power units, it realizes on-demand dynamic combination and precise task scheduling, breaks the fixed pattern of computing power type, significantly improves resource utilization, and relies on the independent management module of each unit to carry out hardware-level real-time power consumption monitoring and scheduling, which can realize deep sleep and fast wake-up at the module level. It breaks through the limitations of the lagging and coarse granularity of traditional OS layer energy saving solutions, and achieves refined real-time optimization of platform-level energy efficiency.

[0031] 2. In this utility model, the computing power processing equipment is mainly composed of a general computing unit, a general service unit, and an AI intelligent computing unit. Through the modular and hot-swappable design of LRM, the modules can be flexibly configured according to the user's needs and computing scenarios to realize the comprehensive processing of data and signals of multiple LRM business modules of graphics, data and signals, providing high-capacity and high-communication information services. When a single module fails, it can be hot-replaced, the fault recovery speed is improved, and the redundancy, security and reliability of the equipment's services are enhanced. Attached Figure Description

[0032] Figure 1This is a three-dimensional schematic diagram of a high-performance computing device proposed in this utility model;

[0033] Figure 2 This is a partial structural diagram of the movable handle of a high-computing-power processing device proposed in this utility model;

[0034] Figure 3 This is a schematic diagram of the general computing unit composition of a high-performance computing device proposed in this utility model;

[0035] Figure 4 This is a schematic diagram illustrating the composition of a general-purpose service unit for a high-computing-power processing device proposed in this utility model.

[0036] Figure 5 This is a schematic diagram of the AI ​​intelligent computing unit composition of a high-computing-power processing device proposed in this utility model.

[0037] Legend:

[0038] 1. Housing; 2. LRM connector; 3. External interface; 4. Debugging and maintenance interface; 5. Power switch; 6. Decryption switch; 7. Movable handle. Detailed Implementation

[0039] The technical solutions of the present utility model will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present utility model, and not all embodiments. Based on the embodiments of the present utility model, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present utility model.

[0040] Reference Figures 1-2 The present invention provides an embodiment of a high-performance computing device, comprising three boxes 1, each encapsulating a general computing unit, a general service unit, and an AI intelligent computing unit. The rear panel of each box 1 is provided with an LRM connector 2 and an external interface 3. The front panel of each box 1 is provided with a movable handle 7 and a debugging and maintenance interface 4, a power switch 5, and a password destruction switch 6. The password destruction switch 6 is a fully sealed self-resetting button switch.

[0041] Specifically, the device is based on an LRM architecture and consists of a general-purpose computing unit, an AI intelligent computing unit, and a general-purpose service unit. A movable handle 7 is installed on the front panel of the housing assembly, allowing for easy module insertion and removal. A lock-on bolt is installed at the top of the front panel of housing 1 for easy tightening after installation. The overall design adopts an LRM architecture with a solderless design and is made of stainless steel. A power switch 5 is used to power on and off each unit. Each unit's carrier board integrates an internal connection terminal of an LRM connector 2. This internal connection terminal is structurally fixed to the LRM connector 2 of housing 1 and connected to the circuit. Each unit is mainly composed of the housing... The device consists of three parts: components, end caps, and carrier board components. Each box 1 can be individually packaged into a unit, enabling it to connect to the backplane independently and achieve a modular effect. Through modular design and heat dissipation simulation, it can achieve functions such as quick plug-in and unplugging of each module, safe locking, solderless assembly, and complete modularity. At the same time, users can plug and unplug computing modules as needed, adapt to scenarios, and flexibly configure according to computing scenarios to realize the comprehensive processing of data and signals from multiple LRM service modules of graphics, data, and signals, providing high-capacity and high-communication information services. In the event of a single module failure, it can be hot-swapped, and the fault recovery speed is improved, enhancing the redundancy, security, and reliability of the equipment's services.

[0042] Reference Figure 3 The general-purpose computing unit includes a general-purpose computing unit carrier board, a storage unit, a management module, and a GPU module. The general-purpose computing unit carrier board integrates LRM connector 2, CPU, memory, bridge chip, network controller, CAN transceiver, and serial transceiver. The storage unit includes a system disk and a data disk.

[0043] Specifically, the general-purpose computing unit carrier board includes processor and bridge interconnect design, memory interconnect, bus design, and embedded control. The general-purpose computing unit processor is mainly based on the Phytium D2000 8-core processor, used to complete general-purpose parallel computing and business processing logic. The processor is interconnected with the X100 IO expansion chip processor through the PCIe design bus to avoid x86 architecture dependency. The X100 then takes over high-speed peripherals, freeing up CPU computing power and avoiding I / O interrupts from occupying computing resources. At the same time, it is interconnected with memory through two channels of DQ bus, and interconnected with network controller through PCIe bus. The processor also has an external SPI Flash connected to it through QSPI bus. The ROM chip stores BIOS code. The processor connects to the EC chip via the LPC bus to control the power-on / off sequence, as well as the hibernation and sleep sequences of the carrier board. The processor can also output three UART serial port signals. The GPU is primarily based on the Jingjia Micro JM7500, capable of efficiently performing 2D and 3D graphics acceleration functions to meet graphics processing requirements. The management module can monitor the voltage and temperature of the general-purpose slots in real time, and determine whether an alarm is triggered by indicator lights. Alarm information is also reported to the main control unit for display. Furthermore, the management module supports unified management of high-performance processing units. It possesses intelligent management functions. The management module communicates with the main control unit via the IPMB bus, and can also perform power-on and power-off operations on the general-purpose slots, and read the power-on and power-off status of the general-purpose slots. At the bus design level, a peripheral domain is constructed through three heterogeneous buses: CAN-FD, RS-485, and Gigabit Ethernet, to ensure that critical commands can still be completed in millisecond-level response even in environments with strong electromagnetic interference. The system disk and data disk are used for system operation and business data storage, respectively. All components are interconnected through standard interfaces, and finally, high-speed interconnection with the backplane is achieved through LRM connectors to realize hot-swappable and modular operation and maintenance.

[0044] Reference Figure 4 The general service unit includes a general service unit carrier board, a storage unit, a management module, and a GPU module. The general service unit carrier board integrates an LRM connector 2, a KP920 module, a BMC module, a network controller, a CAN transceiver, and a serial transceiver. The storage unit includes a system disk and a data disk. The general service unit carrier board includes a network unit design and a bus design.

[0045] Specifically, the CPU of the general-purpose service unit is mainly based on the high-performance processor KP920 module, targeting scenarios such as high-performance computing, AI inference, and edge embedding. It integrates high-speed network cards 8×25G / 4×50G, 2×100G, and PCIe 4.0 x32 to meet different expansion needs. The GPU of the general-purpose service unit also chooses Jingjia Micro JM7500, which can efficiently complete 2D and 3D graphics acceleration functions to meet graphics computing requirements. The KP920 module does not have a built-in GPU unit and requires an external GPU unit. The KP920 module acts as the main control core, scheduling data from the storage unit to memory for processing through the high-speed bus on the carrier board, and relying on the integrated high-speed network card to achieve high-throughput data exchange. According to the design roadmap of this device, the display unit is used for debugging and is not a core unit. The BMC module that comes standard with HWKP920 is selected as the display unit, which has the characteristics of low power consumption, small size, and good PCIe device compatibility. In addition to power management and fault monitoring, it integrates a GPU. VGA display functionality: PCIe x1 outputs VGA via module conversion. The general-purpose service unit carrier board uses the KP920 high-performance processor module as its computing center. It integrates a high-speed network controller through optimized network unit design and utilizes an advanced bus design to achieve high-speed data exchange with various components. The general-purpose service unit carrier board integrates a BMC module responsible for real-time hardware status monitoring, power management, and remote control functions. It also features a CAN transceiver and a serial transceiver for communication with external devices. The system disk and data disk respectively host the operating system and business data storage, achieving fine-grained module-level management through a management module, and integrate a GPU module for graphics acceleration. With its computing capabilities, the GPU module accelerates graphics or computing tasks in parallel by receiving them from the main processor. All components work together through the high-speed interconnect bus inside the carrier board and ultimately interface with the system backplane via the standardized LRM connector 2, supporting hot-swapping and flexible expansion. The BMC module independently monitors the module voltage, temperature, and operating status and uploads them to the chassis management unit via the management bus. External device data can be accessed via CAN / serial port and transmitted to the main processor via the carrier board bus. All business flows, management flows, and data flows are converged and distributed within the carrier board and finally interconnected with the backplane via the LRM connector, realizing modular service processing and unified resource pooling.

[0046] Reference Figure 5 The AI ​​intelligent computing unit includes an AI intelligent computing unit carrier board and a management module. The AI ​​intelligent computing unit carrier board integrates two LRM connectors, two 310P modules, a BMC module, an FPGA module, a serial transceiver, a network transceiver, and a CAN transceiver. The AI ​​intelligent computing unit outputs six LiDARs, three SDI cameras, six video signals, two RS232 interfaces, one RS422 interface, nine CAN interfaces, and two USB interfaces.

[0047] Specifically, the AI ​​intelligent computing unit carrier board includes processor design, video decoding design, and FPGA design. The AI ​​intelligent computing unit has dedicated output interfaces such as radar interface, camera interface, and video interface. Through the AI ​​intelligent computing unit, it can encode and decode LiDAR signals, video signals, camera signals, etc., and perform high-performance computing, realizing the decoding processing of external input signals. The central processing unit chip of the AI ​​intelligent computing unit uses the Ascend 310P, based on the Da Vinci architecture, supporting up to 140 TOPs of AI computing power INT8, and has half-precision FP16 computing power. Simultaneously, the AI ​​intelligent computing unit supports 128 channels of 1080P@30fps H.264 and H.265 video encoding and decoding, JPEG image encoding and decoding, and 24 channels of hardware encoding of the same specifications, suitable for application scenarios requiring real-time video processing. After deserializing externally received radar signals, video signals, camera signals, etc., into parallel data through the interfaces, it is input to the FPGA for decoding processing, aggregation, and protocol conversion, and then processed through different PCIe interfaces. Data is distributed to the NPU for computation. The AI ​​computing unit LRM module integrates a Phytium D2000 processor and at least two Ascend 310P accelerator cards. The two Ascend 310P cards are directly connected to the Phytium D2000 via PCIe x8 interfaces, providing up to 32Gbps bandwidth to ensure high-speed data transmission. The D2000's DDR4 memory controller serves as a shared cache, storing preprocessed data and intermediate results for direct access by the Ascend 310P, reducing data overhead. The D2000 is responsible for general computing and scheduling, including data preprocessing. Task scheduling and load balancing, system control and I / O management, etc., divide the video stream into segments such as multiple camera inputs and distribute them in parallel to the 310P for processing. The 310P is responsible for neural network inference, video data operation, real-time data analysis, etc., releasing the 140 TOPS@INT8 computing power of the Ascend 310P to focus on neural network inference, with linear improvement in computing power and significant reduction in end-to-end latency. It takes into account the requirements of independent controllability and high performance and low power consumption. The AI ​​intelligent computing unit can complete the encoding and decoding of video signals such as LiDAR signals and SDI signals and perform high-performance computing.

[0048] Reference Figures 2-5 The LRM connector 2 has built-in 10 Gigabit Ethernet interface, Gigabit Ethernet interface, CAN interface, and TSN management interface for high-speed data transmission and management between units; the general computing unit, general service unit, and AI intelligent computing unit are all electrically connected to the password destroyer switch 6.

[0049] Specifically, the general computing unit, general service unit, and AI intelligent computing unit are electrically connected to the high-speed backplane through the standard LRM interface, realizing modular design and hot-swappable functionality. This high-performance computing module is connected to the backplane through LRM connector 2, and then outputs from the backplane to 10 Gigabit, Gigabit, and TSN aviation connectors. Data interaction between external and internal CAN devices is achieved through LRM connector 2. The Ethernet bus provides capabilities for service processing and data transmission exchange. The management bus is used for the management and configuration of each unit within the rack. The TSN bus ensures high-quality end-to-end transmission of critical data streams in mixed traffic environments. Each unit supports key destruction functionality and has a one-click logical data destruction function. The key destruction switch 6 is a fully sealed self-reset button switch.

[0050] Reference Figures 2-5 The general computing unit, general service unit, and AI intelligent computing unit are all electrically connected to power switch 5. Each of the general computing unit, general service unit, and AI intelligent computing unit is equipped with an aviation connector. Each of the general computing unit, general service unit, and AI intelligent computing unit is equipped with an independent status indicator light to indicate the power supply status and operating status of each unit.

[0051] Specifically, the power switch 5 allows the general computing unit, general service unit, and AI intelligent computing unit to be turned on and off individually, and indicator lights can indicate the power supply status and working status of each unit.

[0052] Working Principle: When needed, box 1 is connected to the high-speed backplane of the chassis via LRM connector 2. Based on the LRM architecture, the general-purpose computing unit, AI intelligent computing unit, and general-purpose service unit are respectively housed in three boxes 1, which are hot-swappable independent modules. Data interaction between external and internal CAN devices is achieved through LRM connector 2. Hardware resource pooling and heterogeneous collaboration can be realized through the high-speed backplane. The CPU of the general-purpose computing unit is based on the Phytium D2000 8-core processor, which is responsible for main control scheduling and general computing tasks. The central processing unit chip of the AI ​​intelligent computing unit is the Ascend 310P, based on the Da Vinci architecture, which is dedicated to neural network inference and video encoding and decoding via PCIe direct connection. The CPU of the general-purpose service unit is based on the high-performance processor KP920 module, which provides high-speed network and service processing. Each module exchanges data through the backplane Ethernet and TSN bus, and the independent management module of each module communicates with the main control unit through the IPMB bus to realize hardware-level status monitoring and power consumption scheduling, and to uniformly manage the high-computing processing units. It can realize the opening of general-purpose slots. The device enables power-on and power-off operations, while simultaneously reading the power-on / off status of general-purpose slots. Ultimately, through dynamic load allocation and module-level energy efficiency control, it achieves flexible configuration and optimal energy efficiency of heterogeneous computing power. The device physically decouples general-purpose computing, AI computing, and service processing units through an LRM modular architecture. Relying on a high-speed backplane, it achieves heterogeneous computing power resource pooling and hardware-level dynamic scheduling, effectively solving the problem of fixed computing power types. At the same time, through independent management modules of each unit, it monitors the power consumption status in real time and achieves module-level deep sleep and fast wake-up hardware energy efficiency control via the IPMB bus. This can improve resource utilization while achieving refined real-time energy efficiency optimization.

[0053] Finally, it should be noted that the above description is only a preferred embodiment of the present utility model and is not intended to limit the present utility model. Although the present utility model has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present utility model should be included within the protection scope of the present utility model.

Claims

1. A high-performance computing device, comprising three housings (1), characterized in that: The three boxes (1) respectively encapsulate a general computing unit, a general service unit and an AI intelligent computing unit. The rear panel of the box (1) is provided with an LRM connector (2) and an external interface (3). The front panel of the box (1) is equipped with a movable handle (7). The front panel of the box (1) is provided with a debugging and maintenance interface (4), a power switch (5) and a password destruction switch (6). The password destruction switch (6) is a fully sealed self-resetting button switch.

2. The high-performance computing device according to claim 1, characterized in that: The general-purpose computing unit includes a general-purpose computing unit carrier board, a storage unit, a management module and a GPU module. The general-purpose computing unit carrier board integrates an LRM connector (2), a CPU, memory, a bridge chip, a network controller, a CAN transceiver and a serial transceiver. The storage unit includes a system disk and a data disk.

3. The high-performance computing device according to claim 1, characterized in that: The general service unit includes a general service unit carrier board, a storage unit, a management module and a GPU module. The general service unit carrier board integrates an LRM connector (2), a KP920 module, a BMC module, a network controller, a CAN transceiver and a serial transceiver. The storage unit includes a system disk and a data disk. The general service unit carrier board includes a network unit design and a bus design.

4. The high-performance computing device according to claim 1, characterized in that: The AI ​​intelligent computing unit includes an AI intelligent computing unit carrier board and a management module. The AI ​​intelligent computing unit carrier board integrates an LRM connector (2), 310P module ×2, BMC module, FPGA module, serial transceiver, network transceiver and CAN transceiver. The AI ​​intelligent computing unit outputs 6 LiDARs, 3 SDI cameras, 6 video signals, 2 RS232s, 1 RS422, 9 CAN interfaces and 2 USB interfaces.

5. The high-performance computing device according to claim 1, characterized in that: The LRM connector (2) has built-in signals including a 10 Gigabit Ethernet interface, a Gigabit Ethernet interface, a CAN interface, and a TSN management interface, which are used for high-speed data transmission and management between units.

6. The high-performance computing device according to claim 1, characterized in that: The general computing unit, general service unit and AI intelligent computing unit are all electrically connected to the secret-breaking switch (6).

7. The high-performance computing device according to claim 1, characterized in that: The general computing unit, general service unit and AI intelligent computing unit are all electrically connected to the power switch (5), and each of the general computing unit, general service unit and AI intelligent computing unit is equipped with an aviation connector.

8. The high-performance computing device according to claim 1, characterized in that: The general computing unit, general service unit, and AI intelligent computing unit are each equipped with an independent status indicator light to indicate the power supply status and operating status of their respective units.

Images