An immersion liquid-cooled edge AI computing server

Abstract

The present invention discloses an immersion-type liquid-cooled edge AI computing power server, comprising: a heat dissipation system, a sensor system, a control system, and a digital native system. The heat dissipation system comprises: an integrated sealed chassis shell, a closed liquid cooling chamber, coolant, heat dissipation fins and a liquid cooling pump; the sensor system comprises a temperature sensor module and a pressure sensor module; the control system comprises a data acquisition module, a first judgment module, a second judgment module and a third judgment module; the digital native system comprises: a zero-trust configuration module, a network configuration module, a storage management module and a virtualization configuration module. The present invention uses a heat dissipation system composed of liquid cooling conduction and heat dissipation fins, so that the server operation does not rely on an external computer room, and can be deployed in non-standard environments such as edge computing nodes and unmanned stations, to achieve plug-and-play AI computing power output, without relying on professional deployment and operation and maintenance, and improve reliability while lowering the deployment threshold.

Description

Technical Field

[0001] The present invention relates to the field of edge computing servers, and in particular to an immersion liquid-cooled edge AI computing server. Background Art

[0002] With the development of artificial intelligence, large-model training, and edge computing, the demand for high-performance computing equipment is growing. Traditional AI servers are usually deployed in standard data centers, relying on specific computer room humidity and temperature environments, as well as professional equipment hardware maintenance personnel, IT personnel, and complex air conditioning and power systems. The energy consumption required for server operation is high, the daily maintenance costs are high, and it is difficult to promote them to edge or unmanned scenarios.

[0003] AI computing power reasoning and training have put forward requirements for servers to reduce energy consumption and increase computing power density. Therefore, liquid-cooled servers have been rapidly accepted by the market due to their high heat dissipation efficiency, low noise, and freedom from geographical restrictions, becoming the mainstream development direction of servers. However, current liquid cooling solutions mostly rely on professional deployment and operation and maintenance, and still have problems with high deployment barriers and weak reliability.

[0004] Whether it is an air-cooled server or a liquid-cooled server, users need to perform network configuration, operating system installation, virtualization and other IT operations on the server after purchasing it so that the server can run the user's business service calculations. This method is time-consuming and has high labor costs.

[0005] Therefore, there is an urgent need for a hardware and software integrated device that can be deployed in non-standard computer room environments, has a high degree of self-perception and self-operation and maintenance capabilities, and is convenient for users to allocate resources and deploy and install business services, so as to realize a truly plug-and-play, unmanned maintenance AI computing infrastructure. Summary of the Invention

[0006] The purpose of this invention is to provide an immersion liquid-cooled edge AI computing power server, which aims to solve the problem that traditional AI servers are usually deployed in standard data centers, rely on specific computer room humidity and temperature environments and professional equipment maintenance personnel and IT personnel, and have high heat generation and high energy consumption, high deployment, configuration and maintenance costs, and are difficult to promote to edge machine room-less scenarios.

[0007] The present application discloses an immersion-type liquid-cooled edge AI computing server, including a heat dissipation system, a sensor system, a control system, and a digital native system, wherein:

[0008] The cooling system includes:

[0009] One-piece sealed chassis shell made of aluminum alloy;

[0010] A closed liquid cooling chamber is formed inside the integrated sealed chassis shell;

[0011] The closed liquid cooling chamber is filled with coolant, which seamlessly immerses the server components, including the motherboard, memory, network card, CPU, GPU, disk, and power supply.

[0012] An integrated sealed heat sink fin on the outer surface of the bottom of the chassis shell;

[0013] Integrated sealed chassis housing with heat dissipation fins on the inner and outer sides of the upper cover;

[0014] A liquid cooling pump connected to the top outlet and the bottom outlet of the liquid cooling chamber is used to form convection of the coolant in the liquid cooling chamber and disperse the heat to all the coolant in the liquid cooling chamber;

[0015] The sensor system includes:

[0016] Temperature sensor module, used to collect the temperature of the top outlet and bottom outlet of the liquid cooling chamber, CPU and GPU;

[0017] The pressure sensor module is used to sense the pressure of the coolant in the liquid cooling chamber and obtain the coolant height;

[0018] The control system includes:

[0019] The data acquisition module is used to create a data acquisition thread and read the data collected by the sensor system at a preset interval;

[0020] a first determination module, configured to determine that there is a coolant leak when the coolant height obtained by the pressure sensor module is lower than a preset height warning value;

[0021] The second determination module is configured to issue a server overheating warning when the temperature at the bottom outlet of the liquid cooling chamber collected by the temperature sensor module exceeds a preset temperature warning value, and simultaneously initiate an instruction to execute load balancing;

[0022] a third determination module, configured to control the rotation speed of the liquid cooling pump according to a difference between an average value of all CPU temperatures collected by the temperature sensor module and a CPU temperature warning value, or according to a difference between a maximum value of all GPU temperatures collected by the temperature sensor module and a GPU temperature warning value;

[0023] Digital native systems include:

[0024] Zero Trust Configuration module, used to turn off the server power manual switch and disable the server USB interface;

[0025] The network configuration module is used to set the network segment and mask of the server management plane and the data plane, and test the network connectivity;

[0026] A storage management module is used to configure the storage structure according to the attributes of the server;

[0027] The virtualization configuration module is used to perform virtualization configuration on the server according to the virtual computing power mode required by the user.

[0028] Preferably, the direction of the heat dissipation fins on the outer surface of the bottom of the integrated sealed chassis shell is consistent with the wind direction of the rack.

[0029] Preferably, the integrated sealed chassis shell has heat dissipation fins on the inner and outer sides of the upper cover, the inner heat dissipation fins are inserted into the coolant, and the direction of the outer heat dissipation fins is consistent with the heat dissipation fins on the outer surface of the bottom of the integrated sealed chassis shell.

[0030] Preferably, the liquid cooling pump connecting the top outlet and the bottom outlet of the liquid cooling chamber:

[0031] When the temperature difference between the coolant at the top and bottom of the liquid cooling chamber does not reach the preset threshold, the liquid cooling pump does not start, and the heat is dispersed to all the coolant in the liquid cooling chamber through natural convection;

[0032] When the temperature difference between the coolant at the top and bottom of the liquid cooling chamber reaches a preset threshold, the liquid cooling pump starts, causing the coolant in the liquid cooling chamber to form forced convection, dispersing heat to all the coolant in the liquid cooling chamber.

[0033] Preferably, the load balancing specifically includes the following steps:

[0034] If the current server is a stand-alone node and the services provided by the current server have priority classification, the low-priority services will be interrupted; if the services provided by the current server do not have priority classification, a server overheating warning will be issued;

[0035] If the current server is a cluster member, the status of other members in the cluster is queried. After filtering out members with server overheat warnings, a check is made to see if there is a target member with both CPU and GPU loads below 50%. If no target member exists, a server overheat warning is issued. If a target member exists, the target member's information is recorded, and service migration is initiated to migrate the services running on the current server to the target member.

[0036] Preferably, the service migration specifically includes the following steps:

[0037] The target member starts the service to be migrated;

[0038] Add the target member's information and the service to be migrated to the routing target receiving object table;

[0039] The gateway receives the object table based on the routing target, intercepts the request to call the service to be migrated, and forwards it to the target member.

[0040] Preferably, the determination method of the third determination module is:

[0041] If the difference between the average value of all CPU temperatures collected by the temperature sensor module and the CPU temperature warning value, or the difference between the maximum value of all GPU temperatures collected by the temperature sensor module and the GPU temperature warning value, is positive and greater than a preset trigger value, the speed of the liquid cooling pump is increased to a preset high-temperature speed;

[0042] If the speed of the liquid cooling pump is the preset high-temperature speed, and the difference between the average value of all CPU temperatures collected by the temperature sensor module and the CPU temperature warning value, as well as the difference between the maximum value of all GPU temperatures collected by the temperature sensor module and the GPU temperature warning value, are both less than the preset trigger value, the speed of the liquid cooling pump is restored to the normal speed.

[0043] Preferably, the storage management method of the storage management module is:

[0044] If the current server is a management node in the computing resource network, the storage space of the current server is divided into a system disk and a data disk;

[0045] If the current server is a computing node in the computing resource network, the storage space of the current server is added as a member to the existing shared file system;

[0046] If the current server and other servers together constitute a management node or computing node in the computing resource network, the storage space of all servers is constructed into a shared distributed file system.

[0047] Preferably, the configuration method of the virtualization configuration module is:

[0048] Preset the virtualization ratio for CPU and GPU based on the user's super-resolution ratio requirements;

[0049] Get the number of CPU cores and threads, and generate the number of virtual CPUs after super-resolution. The number of virtual CPUs = (actual number of CPU cores × number of threads - number of system loss cores) × 3;

[0050] If the resources required by the user are virtual machines, and the current server is a stand-alone management node or computing power node in the resource pool, the computing resources of the current server will be divided into two parts, one of which will be installed with Kubernetes for deploying digital native system services, and the other will be used to configure virtual machines; if the resources required by the user are containers, all the server resources will be used to install Kubernetes.

[0051] The present invention discloses an immersion-type liquid-cooled edge AI computing server, which has the following beneficial effects:

[0052] (1) The heat dissipation system composed of liquid cooling conduction and heat dissipation fins makes the server operation independent of the external computer room. It can be deployed in non-standard environments such as edge computing nodes and unmanned stations, achieving plug-and-play AI computing power output without relying on professional deployment and operation and maintenance, thereby lowering the deployment threshold and improving reliability.

[0053] (2) It has preset network configuration, storage configuration, and virtualization configuration modules, and has zero IT deployment capabilities, requiring no professional personnel for debugging or deployment.

[0054] (3) Through the control system, the server has the ability to self-diagnose status and self-recover from abnormalities.

[0055] (4) The whole system is highly integrated, with integrated liquid cooling, low noise, energy saving and environmental protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0057] Figure 1 This is the overall structure diagram of an immersion liquid-cooled edge AI computing server. DETAILED DESCRIPTION

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0059] Example:

[0060] like Figure 1 As shown, an embodiment of the present application provides an immersion liquid-cooled edge AI computing server, including a heat dissipation system, a sensor system, a control system, and a digital native system.

[0061] The heat dissipation system includes:

[0062] The one-piece sealed chassis shell is made of aluminum alloy and is manufactured through an extrusion process.

[0063] An enclosed liquid cooling chamber is formed inside the integrated sealed chassis shell.

[0064] The sealed liquid cooling chamber is filled with coolant, which seamlessly immerses server components, including the motherboard, memory, network card, CPU, GPU, disk, and power supply.

[0065] The heat dissipation fins on the outer bottom surface of the integrated sealed chassis shell are aligned with the wind direction of the rack.

[0066] The heat dissipation fins on the inner and outer sides of the upper cover of the integrated sealed chassis shell are inserted into the coolant, and the direction of the heat dissipation fins on the outer surface of the bottom of the integrated sealed chassis shell is consistent with that of the heat dissipation fins on the outer surface of the bottom of the integrated sealed chassis shell.

[0067] The liquid cooling pump connected to the top outlet and the bottom outlet of the liquid cooling chamber is used to make the coolant in the liquid cooling chamber form convection and disperse the heat to all the coolant in the liquid cooling chamber.

[0068] When the temperature difference between the coolant at the top and bottom of the liquid cooling chamber does not reach the preset threshold, the liquid cooling pump does not start, and the heat is dispersed to all the coolant in the liquid cooling chamber through natural convection;

[0069] When the temperature difference between the coolant at the top and bottom of the liquid cooling chamber reaches a preset threshold, the liquid cooling pump starts, causing the coolant in the liquid cooling chamber to form forced convection, dispersing heat to all the coolant in the liquid cooling chamber.

[0070] The sensor system includes:

[0071] The temperature sensor module is used to collect the temperatures of the top and bottom outlets of the liquid cooling chamber, the CPU, and the GPU.

[0072] The pressure sensor module is used to obtain the coolant height by sensing the pressure of the coolant in the liquid cooling chamber.

[0073] The control system includes:

[0074] The data acquisition module is used to create a data acquisition thread and read the data collected by the sensor system at preset intervals.

[0075] The first determination module is configured to determine that there is a coolant leak when the coolant height obtained by the pressure sensor module is lower than a preset height warning value.

[0076] The second determination module is used to issue a server overheat warning when the temperature at the bottom outlet of the liquid cooling chamber detected by the temperature sensor module exceeds a preset temperature warning value, and simultaneously initiate an instruction to execute load balancing. The load balancing specifically includes the following steps:

[0077] If the current server is a stand-alone node and the services provided by the current server have priority classification, the low-priority services will be interrupted; if the services provided by the current server do not have priority classification, a server overheating warning will be issued;

[0078] If the current server is a cluster member, the system queries the status of other members in the cluster, filters out members with server overheat warnings, and then checks whether there is a target member with both CPU and GPU loads below 50%. If no target member exists, a server overheat warning is issued. If a target member exists, the target member's information is recorded and service migration is initiated, migrating the services running on the current server to the target member. The service migration process specifically includes the following steps:

[0079] The target member starts the service to be migrated;

[0080] Add the target member's information and the service to be migrated to the routing target receiving object table;

[0081] The gateway receives the object table based on the routing target, intercepts the request to call the service to be migrated, and forwards it to the target member.

[0082] The third determination module is configured to control the speed of the liquid cooling pump based on the difference between the average value of all CPU temperatures collected by the temperature sensor module and the CPU temperature warning value, or based on the difference between the maximum value of all GPU temperatures collected by the temperature sensor module and the GPU temperature warning value. The specific determination method is as follows:

[0083] If the difference between the average value of all CPU temperatures collected by the temperature sensor module and the CPU temperature warning value, or the difference between the maximum value of all GPU temperatures collected by the temperature sensor module and the GPU temperature warning value, is positive and greater than a preset trigger value, the speed of the liquid cooling pump is increased to a preset high-temperature speed;

[0084] If the speed of the liquid cooling pump is the preset high-temperature speed, and the difference between the average value of all CPU temperatures collected by the temperature sensor module and the CPU temperature warning value, as well as the difference between the maximum value of all GPU temperatures collected by the temperature sensor module and the GPU temperature warning value, are both less than the preset trigger value, the speed of the liquid cooling pump is restored to the normal speed.

[0085] Digital native systems include:

[0086] Zero Trust Configuration module, used to turn off the server power manual switch and disable the server USB interface.

[0087] The network configuration module is used to set the server management plane network segment and mask as well as the data plane network segment and mask, and test network connectivity.

[0088] The storage management module is used to configure the storage structure according to the server's attributes. The specific storage management method is:

[0089] If the current server is a management node in the computing resource network, the storage space of the current server is divided into a system disk and a data disk;

[0090] If the current server is a computing node in the computing resource network, the storage space of the current server is added as a member to the existing shared file system;

[0091] If the current server and other servers together constitute a management node or computing node in the computing resource network, the storage space of all servers is constructed into a shared distributed file system.

[0092] The virtualization configuration module is used to configure the server virtualization according to the virtual computing power mode required by the user. The specific configuration method is:

[0093] Preset the virtualization ratio for CPU and GPU based on the user's super-resolution ratio requirements;

[0094] Get the number of CPU cores and threads, and generate the number of virtual CPUs after super-resolution. The number of virtual CPUs = (actual number of CPU cores × number of threads - number of system loss cores) × 3;

[0095] If the resources required by the user are virtual machines, and the current server is a stand-alone management node or computing node in the resource pool, the computing resources of the current server will be divided into two parts, one of which will be installed with Kubernetes for deploying digital native platform services, and the other will be used to configure virtual machines; if the resources required by the user are containers, all the server resources will be used to install Kubernetes.

[0096] In summary, the immersion liquid-cooled edge AI computing server provided by the embodiments of the present application has the following technical effects:

[0097] (1) The heat dissipation system composed of liquid cooling conduction and heat dissipation fins makes the server operation independent of the external computer room. It can be deployed in non-standard environments such as edge computing nodes and unmanned stations, achieving plug-and-play AI computing power output without relying on professional deployment and operation and maintenance, thereby lowering the deployment threshold and improving reliability.

[0098] (2) It has preset network configuration, storage configuration, and virtualization configuration modules, and has zero IT deployment capabilities, requiring no professional personnel for debugging or deployment.

[0099] (3) Through the control system, the server has the ability to self-diagnose status and self-recover from abnormalities.

[0100] (4) The whole system is highly integrated, with integrated liquid cooling, low noise, energy saving and environmental protection.

[0101] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0102] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An immersion liquid-cooled edge AI computing server, characterized in that: Including cooling system, sensor system, control system, and digital native system, including: The cooling system includes: One-piece sealed chassis shell made of aluminum alloy; A closed liquid cooling chamber is formed inside the integrated sealed chassis shell; The closed liquid cooling chamber is filled with coolant, which seamlessly immerses the server components, including the motherboard, memory, network card, CPU, GPU, disk, and power supply. The heat dissipation fins on the outer surface of the bottom of the integrated sealed chassis shell are oriented in the same direction as the wind direction of the rack; The heat dissipation fins on the inner and outer sides of the upper cover of the integrated sealed chassis shell are arranged such that the heat dissipation fins on the inner and outer sides of the upper cover of the integrated sealed chassis shell are inserted into the coolant, and the heat dissipation fins on the outer side are arranged in the same direction as the heat dissipation fins on the outer surface of the bottom of the integrated sealed chassis shell; A liquid cooling pump connected to the top outlet and the bottom outlet of the liquid cooling chamber is used to form convection of the coolant in the liquid cooling chamber and disperse the heat to all the coolant in the liquid cooling chamber; The liquid cooling pump connecting the top outlet and the bottom outlet of the liquid cooling chamber: When the temperature difference between the coolant at the top and bottom of the liquid cooling chamber does not reach the preset threshold, the liquid cooling pump does not start, and the heat is dispersed to all the coolant in the liquid cooling chamber through natural convection; When the temperature difference between the coolant at the top and bottom of the liquid cooling chamber reaches a preset threshold, the liquid cooling pump starts, causing the coolant in the liquid cooling chamber to form forced convection, dispersing heat to all the coolant in the liquid cooling chamber; The sensor system includes: Temperature sensor module, used to collect the temperature of the top outlet and bottom outlet of the liquid cooling chamber, CPU and GPU; The pressure sensor module is used to sense the pressure of the coolant in the liquid cooling chamber and obtain the coolant height; The control system includes: The data acquisition module is used to create a data acquisition thread and read the data collected by the sensor system at a preset interval; a first determination module, configured to determine that there is a coolant leak when the coolant height obtained by the pressure sensor module is lower than a preset height warning value; The second determination module is configured to issue a server overheating warning when the temperature at the bottom outlet of the liquid cooling chamber collected by the temperature sensor module exceeds a preset temperature warning value, and simultaneously initiate an instruction to execute load balancing; The load balancing specifically includes the following steps: If the current server is a stand-alone node and the services provided by the current server have priority classification, the low-priority services will be interrupted; if the services provided by the current server do not have priority classification, a server overheating warning will be issued; If the current server is a cluster member, the system queries the status of other members in the cluster, filters out members with server overheat warnings, and then checks whether there is a target member with both CPU and GPU loads below 50%. If no target member exists, a server overheat warning is issued. If a target member exists, the system records the target member's information and initiates service migration, migrating services running on the current server to the target member. a third determination module, configured to control the rotation speed of the liquid cooling pump according to a difference between an average value of all CPU temperatures collected by the temperature sensor module and a CPU temperature warning value, or according to a difference between a maximum value of all GPU temperatures collected by the temperature sensor module and a GPU temperature warning value; Digital native systems include: Zero Trust Configuration module, used to turn off the server power manual switch and disable the server USB interface; The network configuration module is used to set the network segment and mask of the server management plane and the data plane, and test the network connectivity; A storage management module is used to configure the storage structure according to the attributes of the server; The virtualization configuration module is used to configure the server virtualization according to the virtual computing power mode required by the user; The configuration method of the virtualization configuration module is: Preset the virtualization ratio for CPU and GPU based on the user's super-resolution ratio requirements; Get the number of CPU cores and threads, and generate the number of virtual CPUs after super-resolution. The number of virtual CPUs = (actual number of CPU cores × number of threads - number of system loss cores) × 3; If the resources required by the user are virtual machines, and the current server is a stand-alone management node or computing power node in the resource pool, the computing resources of the current server will be divided into two parts, one of which will be installed with Kubernetes for deploying digital native system services, and the other will be used to configure virtual machines; if the resources required by the user are containers, all the server resources will be used to install Kubernetes.

2. The immersion liquid-cooled edge AI computing server according to claim 1, characterized in that: The service migration specifically includes the following steps: The target member starts the service to be migrated; Add the target member's information and the service to be migrated to the routing target receiving object table; The gateway receives the object table based on the routing target, intercepts the request to call the service to be migrated, and forwards it to the target member.

3. The immersion liquid-cooled edge AI computing server according to claim 1, characterized in that: The determination method of the third determination module is: If the difference between the average value of all CPU temperatures collected by the temperature sensor module and the CPU temperature warning value, or the difference between the maximum value of all GPU temperatures collected by the temperature sensor module and the GPU temperature warning value, is positive and greater than a preset trigger value, the speed of the liquid cooling pump is increased to a preset high-temperature speed; If the speed of the liquid cooling pump is the preset high-temperature speed, and the difference between the average value of all CPU temperatures collected by the temperature sensor module and the CPU temperature warning value, as well as the difference between the maximum value of all GPU temperatures collected by the temperature sensor module and the GPU temperature warning value, are both less than the preset trigger value, the speed of the liquid cooling pump is restored to the normal speed.

4. The immersion liquid-cooled edge AI computing server according to claim 1, characterized in that: The storage management method of the storage management module is: If the current server is a management node in the computing resource network, the storage space of the current server is divided into a system disk and a data disk; If the current server is a computing node in the computing resource network, the storage space of the current server is added as a member to the existing shared file system; If the current server and other servers together constitute a management node or computing node in the computing resource network, the storage space of all servers is constructed into a shared distributed file system.

Images