Server, heat dissipation control method for server, and electronic device

Abstract

This application discloses a server, a heat dissipation control method for the server, and an electronic device, relating to the field of server technology. The server includes: multiple accelerator cards, each accelerator card including a turbo fan; one or more network cards; a first air guide shroud located above the one or more network cards to guide airflow to the battery area of ​​the one or more network cards; and a second air guide shroud disposed above at least two accelerator cards and blocking the gap between at least two adjacent accelerator cards to guide airflow to the turbo fans after the system fans are activated. This application effectively solves the problems of high system fan power consumption and poor temperature stability of internal components in GPU servers in related technologies.

Description

Technical Field

[0001] This application relates to the field of server technology, and in particular to a server, a heat dissipation control method for a server, and an electronic device. Background Technology

[0002] Currently, GPU servers generally adopt a "dual-stage cooling system". On the one hand, each GPU card (accelerator card) integrates an independent centrifugal turbo fan, which is controlled in real time by the GPU firmware according to its junction temperature, and is used to directly cool the GPU chip and memory. On the other hand, the server chassis is equipped with multiple sets of system fans, which are uniformly controlled by the Baseboard Management Controller (BMC) to guide the overall airflow, exhaust hot air and cool auxiliary power consumption devices such as BF3 network cards, power modules, and onboard chips.

[0003] However, in existing GPU servers, a large amount of cold air generated by the system fan does not flow through the GPU turbine fan. Instead, it leaks directly into the rear air duct through the gaps on the top of the GPU card or between the cards. This "air leakage" phenomenon leads to a significant reduction in effective cooling airflow, forcing the system fan to operate at a higher speed to compensate for the lack of airflow. This not only increases the energy consumption of the system fan but also affects the temperature stability of the internal components of the server. Summary of the Invention

[0004] This application provides a server, a heat dissipation control method for a server, and an electronic device to at least solve the problems of high system fan power consumption and poor temperature stability of internal components in GPU servers in the related art.

[0005] This application provides a server, including: a plurality of accelerator cards, each accelerator card including a turbofan; one or more network cards; a first air guide shroud located above the one or more network cards to guide airflow to the battery area of ​​the one or more network cards; and a second air guide shroud disposed above at least two accelerator cards and blocking the gap between at least two adjacent accelerator cards to guide airflow to the turbofan after the system fan is started.

[0006] Furthermore, the first air guide shroud includes a first plate, an inclined guide plate, and a second plate connected in sequence. The first plate is located at the air inlet, and the inclined guide plate is set at an obtuse angle to both the first plate and the second plate.

[0007] Furthermore, the first air guide is made of at least one of metal, ABS, PP and silicone foam; or the inclined air guide is made of silicone foam or rubber.

[0008] Furthermore, the server also includes a chassis, with a first air guide shroud located inside the chassis. The chassis includes multiple uprights and crossbeams mounted on the uprights. The first air guide shroud also includes a bent plate, which is located at the end of the first plate away from the inclined air guide plate and is set at an angle to the first plate. The bent plate and the first plate form an overlapping portion, which overlaps the crossbeam.

[0009] Furthermore, the elastic modulus E of the material selected for the first air guide shroud is derived from the first formula: I req =I old +[ΔK bending W L 2 ] / (α E); among them, I req To compensate for the required moment of inertia of a single column, I old To compensate for the moment of inertia of the section of the single front column, W is the width of the chassis, L is the internal height of the chassis, α is the support coefficient of the first air guide shroud and satisfies: 0.1≤α≤1.0, ΔK bending The bending stiffness deficit is derived from the second formula: ΔK bending =(F / δ max )-(F / δ actual F is the total load of the chassis, δ max For the maximum allowable deflection, δ actual The actual deflection; if the elastic modulus E of the material selected for the first air guide is less than the first threshold or within the first range, then the first air guide is selected from the first material; if the elastic modulus E of the material selected for the first air guide is greater than or equal to the first threshold and less than or equal to the second threshold or within the second range, then the first air guide is selected from the second material.

[0010] Furthermore, the server also includes a chassis, with a first air duct located inside the chassis, and multiple accelerator cards arranged at intervals along a preset direction with adjacent accelerator cards being parallel to each other; the preset direction is set at an angle to the air intake direction of the chassis.

[0011] Furthermore, the second air guide shroud includes: a cover plate; and multiple insert plates, which are spaced apart along the length of the cover plate on its lower surface, with each insert plate extending between two adjacent accelerator cards.

[0012] Furthermore, the second air guide cover also includes a support column, which is disposed on the lower surface of the cover plate to support the cover plate.

[0013] Furthermore, the server also includes a chassis, with a first air duct located inside the chassis; the cover includes: a third plate, with multiple insert plates disposed on the lower surface of the third plate; and a surround plate disposed on the upper surface of the third plate, the surround plate contacting the side wall of the chassis to support the side wall.

[0014] Furthermore, the second air guide cover is made of at least one of the following materials: metal, ABS, PP, and silicone foam; and / or, the second air guide cover is a one-piece molded structure.

[0015] Furthermore, the elastic modulus E' of the material selected for the second air guide shroud satisfies the third formula: E'≥(ΔK) W 3 ) / (β t 3 H); where W is the width of the chassis, β is the support coefficient of the second air guide shroud and satisfies: 0.8≤β≤1.2, t is the thickness of the enclosure, H is the height of the chassis, and ΔK is the strength defect value to be compensated and is obtained from the fourth formula: ΔK=(q S) (1 / δ max -1 / δ actual ); q is the uniformly distributed load acting on the side wall of the chassis, S is the effective force-bearing area of ​​the side wall of the chassis, δ max δ is the maximum permissible safe deflection at the center of the chassis sidewall. actual The actual deflection of the side wall of the gearbox under load q when it is not reinforced.

[0016] This application also provides a heat dissipation control method for a server, applied to the aforementioned server. The heat dissipation control method includes a two-stage fan control method, which includes: controlling the start of the turbofan of the accelerator card and / or the system fan of the server according to the relationship between the surface temperature of each accelerator card of the server and a first temperature threshold T1 and a second temperature threshold T2; wherein, the first temperature threshold T1 is less than the second temperature threshold T2.

[0017] Furthermore, the method for controlling the start-up of the turbofan and / or system fan of the accelerator card based on the relationship between the surface temperature of each accelerator card and the first temperature threshold T1 and the second temperature threshold T2 includes: if the surface temperature of each accelerator card continues to rise and reaches the first temperature threshold T1, then the turbofan of the accelerator card is started; if the surface temperature of each accelerator card continues to rise and reaches the second temperature threshold T2, then the system fan is started.

[0018] Furthermore, the method for controlling the start-up of the turbofan and / or system fan of the accelerator card based on the relationship between the surface temperature of each accelerator card and the first temperature threshold T1 and the second temperature threshold T2 also includes: if the surface temperature of each accelerator card continues to drop and drops below the second temperature threshold T2, then the speed of the system fan is reduced; if the surface temperature of each accelerator card continues to drop and drops below the first temperature threshold T1, then the system fan is turned off and / or the speed of the turbofan of the accelerator card is reduced.

[0019] Furthermore, methods for reducing the speed of system fans or turbofans include reducing the speed of system fans or turbofans by no more than 5 to 10% of the rated speed per second.

[0020] Furthermore, during the process of controlling the temperature of all accelerator cards, the temperature control methods for all accelerator cards are executed synchronously.

[0021] Furthermore, the heat dissipation control method also includes a temperature monitoring method for a multi-accelerator card server serial polling architecture. This method includes: obtaining the maximum temperature value T of all accelerator cards within the current polling period K. MAX (K), and record the average temperature T of all accelerator cards in the first two polling cycles. AVG (K-1) and T AVG (K-2); and according to the fifth formula:

[0022] FS(k) = FS act (k-1)+K p (T MAX (K)-T AVG (K-1))+K i (T MAX (K)-T sp )+K d (T MAX (K)-2 T AVG (K-1)+T AVG (K-2) Obtain the pulse width modulation output value FS(k) of the server's system fan within the current polling period K, and control the system fan speed based on the pulse width modulation output value FS(k); where FS act (k-1) represents the pulse width modulation output value of the server's system fan in the previous polling cycle, T sp To set the temperature threshold, K p The first coefficient is 1 ≤ K p ≤20, K i It is the second coefficient and satisfies 0.02≤K i ≤0.50, K d It is the third coefficient and satisfies 0.1≤K d ≤4.0.

[0023] Furthermore, based on the maximum temperature value T of all accelerator cards... MAX (K) and the average temperature T of all accelerator cards in the previous polling cycle AVG The relationship between (K-1) adjusts the pulse width modulation output value FS(k) of the system fan.

[0024] Furthermore, based on the maximum temperature value T of all accelerator cards... MAX (K) and the average temperature T of all accelerator cards in the previous polling cycle AVG The method for adjusting the pulse width modulation output value FS(k) of the system fan based on the relationship between (K-1) includes: if the maximum temperature value T of all accelerator cards... MAX (K) and the average temperature T of all accelerator cards in the previous polling cycle AVG If there is a fixed deviation A between (K-1), then the pulse width modulation output value FS'(k) of the system fan will be adjusted to: FS'(k) = FS(k) + K p A.

[0025] Furthermore, based on the maximum temperature value T of all accelerator cards... MAX (K) and the average temperature T of all accelerator cards in the previous polling cycle AVG The method for adjusting the pulse width modulation output value FS(k) of the system fan based on the relationship between (K-1) includes: if the maximum temperature value T of all accelerator cards... MAX (K) and the average temperature T of all accelerator cards in the previous polling cycle AVG If the values ​​(K-1) are proportional, then the pulse width modulation output value FS'(k) of the system fan will be adjusted to: FS'(k) = FS(k) + K p T MAX (K) (B-1); where B is a proportionality coefficient greater than 1.

[0026] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for implementing the heat dissipation control method for a server as described above when executing the computer program.

[0027] Applying the technical solution of this application, when the server is operating, the system fan exhausts hot air through the rear window of the chassis to create negative pressure, while the front window draws in cool air. A first air guide shroud, located at the front window, directly guides the airflow to the battery area of ​​one or more network cards, ensuring that critical components receive sufficient cool air. Simultaneously, a second air guide shroud is positioned above all accelerator cards and blocks the gaps between adjacent accelerator cards, effectively preventing bypasses of cool air leakage from between the accelerator cards and forcing the airflow through the accelerator cards' built-in turbofans for efficient heat dissipation. In this optimized airflow organization, the system fan does not need to maintain excessively high speeds to compensate for air leakage. This significantly reduces the ineffective energy consumption of the system fan while ensuring the temperature stability of internal components such as accelerator cards and network cards. It solves the problems of high system fan energy consumption and poor temperature stability of internal server components in related technologies for GPU servers, improves the temperature stability of internal server components, avoids frequent acceleration and deceleration of the system fan, and extends the lifespan of the system fan. Attached Figure Description

[0028] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 A three-dimensional structural diagram of the server after its first air guide shroud, chassis, and network card are assembled, as provided in the embodiments of this application.

[0030] Figure 2 for Figure 1 A partial 3D structural diagram of the first air guide shroud after assembly with the chassis and network card;

[0031] Figure 3 for Figure 1 A three-dimensional structural diagram of the first air guide shroud in the middle;

[0032] Figure 4 for Figure 3 A three-dimensional structural diagram of the first air guide shroud from another angle;

[0033] Figure 5 A three-dimensional structural diagram of the server after assembly with the chassis and accelerator card, as provided in the embodiments of this application;

[0034] Figure 6 for Figure 5 A three-dimensional structural diagram of the second air guide shroud.

[0035] The above figures include the following reference numerals:

[0036] 10. Chassis; 11. Front window;

[0037] 20. Accelerator Card;

[0038] 30. Network card; 31. Battery area;

[0039] 40. First air guide shroud; 41. First plate; 42. Inclined guide vane; 43. Second plate; 44. Bending plate;

[0040] 50. Second air guide hood; 51. Cover plate; 511. Third plate; 512. Enclosure plate; 52. Insert plate; 53. Support column. Detailed Implementation

[0041] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0042] It should be noted that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing this application and simplifying the description. They do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. The terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can be fixed connections, detachable connections, or integral connections; they can be mechanical connections or electrical connections; they can be direct connections or indirect connections through an intermediate medium; they can be internal connections between two elements. The terms "parallel," "perpendicular," and "equal" include the described situation and situations similar to the described situation, where the range of similarity is within an acceptable deviation range, which is determined by those skilled in the art taking into account the measurement under discussion and the errors associated with the measurement of a particular quantity, i.e., the limitations of the measurement system. For example, "parallel" includes absolute parallelism and approximate parallelism, where an acceptable deviation range for approximate parallelism can be, for example, within 5°; "perpendicular" includes absolute perpendicularity and approximate perpendicularity, where an acceptable deviation range for approximate perpendicularity can also be, for example, within 5°. "Equal" includes absolute equality and approximate equality, where an acceptable deviation range for approximate equality can be, for example, a difference between the two equal items being less than or equal to 5% of either one. Those skilled in the art will understand the specific meaning of the above terms in this application based on the specific circumstances.

[0043] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0044] To address the issues of high system fan power consumption and poor temperature stability of internal components in GPU servers in related technologies, this application provides a server, a heat dissipation control method for the server, and an electronic device.

[0045] like Figures 1 to 6 As shown, the server includes multiple accelerator cards 20, one or more network cards 30, a first air duct 40, and a second air duct 50. Each accelerator card 20 includes a turbofan. The first air duct 40 is located above one or more network cards 30 to guide airflow to the battery area 31 of one or more network cards 30. The second air duct 50 is disposed above at least two accelerator cards 20 and blocks the gap between at least two adjacent accelerator cards 20 to guide airflow to the turbofan after the system fan is activated.

[0046] Applying the technical solution of this embodiment, when the server is working, the system fan exhausts hot air through the rear window of the chassis 10 to create negative pressure, while the front window draws in cool air. The first air guide shroud 40 is located at the front window and directly guides the airflow to the battery area of ​​one or more network cards 30, ensuring that critical components receive sufficient cool air. Simultaneously, the second air guide shroud 50 is positioned above all the accelerator cards 20 and blocks the gap between adjacent accelerator cards 20, effectively blocking the bypass path of cool air leakage from the gaps between the accelerator cards 20 (GPU cards), forcing the airflow through the turbofans built into the accelerator cards 20 for efficient heat dissipation. Thus, through the optimized airflow organization described above, the system fan does not need to maintain an excessively high speed to compensate for air leakage. This significantly reduces the ineffective energy consumption of the system fan while ensuring the temperature stability of internal components such as the accelerator cards 20 and network cards. It solves the problems of high system fan energy consumption and poor temperature stability of internal server components in related technologies for GPU servers, improves the temperature stability of internal server components, avoids frequent acceleration and deceleration of the system fan, and extends the service life of the system fan.

[0047] In this embodiment, by setting a first air guide shroud 40 on the front window of the chassis and positioning it above the network card 30 to force the intake airflow directly to the network card battery area, combined with the second air guide shroud 50 sealing the airflow at the top and gaps of the GPU card, directional airflow distribution is achieved, avoiding cold air bypass. This stabilizes the network card battery operating temperature within a safe threshold, and the system fan does not need to run at high speed globally to compensate for battery overheating. This solves the technical problem of reliability failures such as battery failure and system crashes caused by insufficient heat dissipation in the network card battery area in existing multi-GPU servers. At the same time, by constructing a "front-supply, rear-sealing" airflow channel with the first air guide shroud 40 and the second air guide shroud 50, the system fan only needs to play an auxiliary role in maintaining overall air pressure and heat dissipation, rather than being the main heat dissipator for the GPU core. Based on the first formula, precise, low-power, and highly stable PWM speed control of the system fan is achieved, reducing the system fan load, improving the response accuracy of the control algorithm, and eliminating sources of airflow interference.

[0048] like Figure 3 and Figure 4As shown, the first air guide shroud 40 includes a first plate 41, an inclined guide plate 42, and a second plate 43 connected in sequence. The first plate 41 is located at the air inlet of the front window 11. The inclined guide plate 42 is set at an obtuse angle to both the first plate 41 and the second plate 43. The first air guide shroud 40 is made of at least one of the following materials: metal, ABS, PP, and silicone foam. Alternatively, the inclined guide plate 42 is made of silicone foam or rubber. By designing the first air guide shroud 40 to be composed of the first plate 41, the inclined guide plate 42, and the second plate 43 connected in sequence, with the inclined guide plate 42 set at an obtuse angle to both the front and rear plates, the purpose of guiding the incoming airflow smoothly, reducing airflow separation and eddy current loss, and improving wind pressure utilization is achieved. This efficiently and centrally guides the cold air to the battery area of ​​the network card 30, improving local heat dissipation efficiency and reducing battery temperature rise. Meanwhile, by using at least one of the following materials to manufacture the first air guide shroud 40: metal, ABS, PP, and silicone foam, and preferably using silicone foam or rubber to make the inclined air guide plate 42, a dynamic balance is achieved between rigid structural support and flexible sealing compensation. This achieves the technical effects of adapting to network cards 30 of different lengths / models, automatically fitting the gaps between the chassis and the network card edges, eliminating air leakage channels, and absorbing mechanical vibration noise.

[0049] like Figure 3 and Figure 4 As shown, the server also includes a chassis 10, with a first air guide shroud 40 located inside the chassis 10. The chassis 10 includes multiple columns and crossbeams mounted on the columns. The first air guide shroud 40 also includes a bent plate 44, which is located at the end of the first plate 41 away from the inclined guide plate 42 and forms an angle with the first plate 41. The bent plate 44 and the first plate 41 form an overlapping portion, which overlaps onto the crossbeams. In this way, the bent plate 44 is provided on the first air guide shroud 40, and it forms an angled structure with the first plate 41, constituting a stable overlapping portion that can overlap onto the chassis crossbeams. This allows the mechanical load of the first air guide shroud 40 to be directly transferred to the main structure of the chassis (columns and crossbeams), thereby significantly improving the overall bending stiffness of the chassis, suppressing sidewall deformation and vibration, and reducing chassis bulging or denting caused by wind pressure negative pressure. Meanwhile, by optimizing the installation method of the first air guide shroud 40 from "point fixing" to a surface contact lap structure along the entire length of the crossbeam (linear lap between the bent plate 44 and the crossbeam), local stress concentration is eliminated, installation tolerance sensitivity is reduced, and assembly consistency and disassembly convenience are improved.

[0050] In this embodiment, the chassis 10 includes a front window 11 and a rear window. Each accelerator card 20 includes a turbofan. A first air guide shroud 40 is located at the front window 11.

[0051] In this embodiment, the elastic modulus E of the material selected for the first air guide shroud 40 is obtained from the first formula:

[0052] I req =I old +[ΔK bending W L 2 ] / (α E);

[0053] Among them, I req To compensate for the required moment of inertia of a single column, I old To compensate for the moment of inertia of the front single column section, W is the width of the chassis 10, L is the inner height of the chassis 10, α is the support coefficient of the first air guide shroud 40 and satisfies: 0.1≤α≤1.0, ΔK bending The bending stiffness deficit is derived from the second formula:

[0054] ΔK bending =(F / δ max )-(F / δ actual F is the total load of chassis 10, δ max For the maximum allowable deflection, δ actual This represents the actual deflection.

[0055] Specifically, based on the bending stiffness deficit ΔK bending Calculate the elastic modulus E of the material selected for the first air guide shroud 40, quantify the structural strength defects of the chassis into input mechanical parameters, and then use I... req The reverse derivation and precise selection of the elastic modulus E of the material selected for the first air guide shroud 40 were achieved, enabling automatic matching of the shroud material stiffness according to actual working conditions (load, deformation), thereby minimizing material usage and avoiding excessive use of steel or misuse of low-modulus plastics. Simultaneously, by dynamically coupling the material selection of the first air guide shroud 40 with the overall structural strength requirements of the chassis, key geometric parameters such as the support coefficient α, width W, and height L included in the formula were used to achieve a functional leap from a "passive air guiding component" to an "active structural reinforcement component." This resulted in a synergistic optimization of three key aspects: overall lightweighting (reduced steel usage), cost reduction (avoiding local thickening), and improved reliability (100% PA test pass rate).

[0056] Optionally, if the elastic modulus E of the material selected by the first air guide shroud 40 is less than the first threshold or within the first range, then the first air guide shroud 40 selects the first material; if the elastic modulus E of the material selected by the first air guide shroud 40 is greater than or equal to the first threshold and less than or equal to the second threshold or within the second range, then the first air guide shroud 40 selects the second material.

[0057] In this embodiment, if the elastic modulus E of the material selected for the first air guide shroud 40 satisfies: E < 69 GPa, then the first air guide shroud 40 is made of aluminum; if the elastic modulus E of the material selected for the first air guide shroud 40 satisfies: 69 GPa < E < 210 GPa, then the first air guide shroud 40 is made of steel. Thus, by using the material's elastic modulus E as the sole criterion and setting 69 GPa and 210 GPa as key dividing thresholds for the air guide shroud material selection, the optimal material type (aluminum or steel) is automatically matched while meeting the chassis's bending stiffness compensation requirements. This achieves the goal of reducing the weight of the first air guide shroud while ensuring structural reliability. Meanwhile, by solidifying material selection decisions into binary logical judgments based on the physical properties of elastic modulus, and combining the calculation of bending stiffness deficit with the design of the first air guide shroud structure, the subjective judgment of human beings in the design process is eliminated, and the purpose of avoiding material misselection (such as using plastic instead of steel or steel instead of aluminum) is avoided. This achieves the technical effect of automatic adaptation and zero-error material selection for air guide shrouds under different chassis specifications (4U, 8U) and different loads (16 cards, 32 cards).

[0058] In this embodiment, multiple accelerator cards 20 are spaced apart along a preset direction and are parallel to each other between adjacent accelerator cards 20; the preset direction is set at an angle to the air intake direction of the chassis 10. Optionally, the preset direction is set at a 90° angle to the air intake direction of the chassis 10.

[0059] like Figure 6 As shown, the second air guide shroud 50 includes a cover plate 51, multiple insert plates 52, and a support column 53. The multiple insert plates 52 are spaced apart along the length of the cover plate 51 on its lower surface, and each insert plate 52 extends into the gap between two adjacent accelerator cards 20. The support column 53 is provided on the lower surface of the cover plate 51 to support the cover plate 51. In this way, by spaced apart multiple insert plates 52 on the lower surface of the cover plate 51 and ensuring that each insert plate 52 extends precisely into the gap between adjacent accelerator cards 20, the bypass channel for cold air to escape from the top of the accelerator card 20 and the gap between the cards is physically blocked, thereby forcibly guiding the airflow of the system fan to the turbofan inlet of the accelerator card 20 and eliminating localized hot spots caused by air leakage. Meanwhile, by setting support columns 53 on the lower surface of the cover plate 51 as load-bearing points, the cover plate 51 can achieve stable suspended support without relying on the chassis beam or the accelerator card 20. It also forms a point and surface collaborative support system with the insert plate 52, eliminating the risk of vibration, deformation and resonance of the cover plate 51 under high wind pressure conditions. This achieves the technical effect of long-term operation of the air guide cover without loosening, abnormal noise, or damage to the top components of the accelerator card 20, and improving compatibility with accelerator cards of different thicknesses.

[0060] like Figure 6As shown, the cover plate 51 includes a third plate 511 and a surrounding plate 512. Multiple insert plates 52 are disposed on the lower surface of the third plate 511. The surrounding plate 512 is disposed on the upper surface of the third plate 511, and contacts the side wall of the chassis 10 to support the side wall of the chassis 10. In this way, the surrounding plate 512 directly contacts and supports the side wall of the chassis 10, making the second air guide shroud an integrated structural load-bearing and side wall reinforcement component, thereby significantly improving the bending stiffness of the chassis side wall, suppressing side wall indentation deformation under high wind pressure conditions, and reducing the overall vibration amplitude. Simultaneously, by physically separating the insert plates 52 from the surrounding plate 512 (structural support function), mutual interference between the airflow channel and the structural load path is eliminated. This ensures that while the insert plates 52 are accurately inserted into the gap of the accelerator card and achieve efficient airtight sealing, the supporting force of the surrounding plate 512 on the chassis side wall is uniformly, stably, and without deformation, transferred to the accelerator card.

[0061] Optionally, the second air guide shroud 50 is made of at least one of the following materials: metal, ABS, PP, and silicone foam; and / or, the second air guide shroud 50 is a one-piece molded structure. This allows the second air guide shroud 50 to be manufactured using at least one of the following materials: metal, ABS, PP, and silicone foam. Material selection is differentiated according to functional areas (e.g., metal or ABS for load-bearing areas, silicone foam for sealing areas), achieving an optimal match between material performance and cost while ensuring structural rigidity, heat resistance, and airtightness. This results in a multi-dimensional functional synergy between the air guide shroud and other components, combining high-strength support (metal / ABS), lightweight design (PP), and active sealing compensation (silicone foam). Furthermore, by designing the second air guide shroud 50 as a one-piece molded structure, potential failure sources such as seams, tolerance accumulation, and fastener failure caused by multi-component assembly are eliminated, thereby improving the structural integrity of the second air guide shroud and enhancing its vibration and fatigue resistance.

[0062] In this embodiment, the elastic modulus E' of the material selected for the second air guide shroud 50 satisfies the third formula:

[0063] E'≥(ΔK W 3 ) / (β t 3 H);

[0064] Where W is the width of the chassis 10, β is the support coefficient of the second air guide shroud 50 and satisfies: 0.8≤β≤1.2, t is the thickness of the enclosure 512, H is the height of the chassis 10, and ΔK is the strength defect value to be compensated and is obtained from the fourth formula:

[0065] ΔK=(q S) (1 / δ max -1 / δactual ); q is the uniformly distributed load acting on the side wall of the chassis 10, S is the effective stress area of the side wall of the chassis 10, and δ max is the maximum allowable safety deflection of the center of the side wall of the chassis 10, and δ actual is the actual deflection of the side wall of the chassis 10 without reinforcement under the load q.

[0066] Specifically, by establishing the actual deflection δ actual of the side wall of the chassis and the allowable maximum deflection δ max as inputs, the strength defect value ΔK is derived, and further, the elastic modulus E' of the material selected for the second air guide cover 50 is used to accurately calculate the elastic modulus E' of the material of the second air guide cover, so that without modifying the main structure of the chassis, the stiffness deficit of the side wall is compensated by the intelligent material selection of the second air guide cover 50, ensuring that the whole machine passes the PA vibration and mechanical tests. At the same time, by using the shroud 512 of the second air guide cover 50 as a structural reinforcement unit and forcing the elastic modulus E' of its material to meet the critical value based on the above formula, the stiffness compensation of the side wall of the chassis is synchronously completed on the premise of ensuring airtightness (insert plate 52) and self-support (support column 53).

[0067] In this embodiment, the present application further provides a heat dissipation control method for a server, which is applied to the above-mentioned server. The heat dissipation control method includes a two-stage fan control method, and the two-stage fan control method includes: controlling the start of the vortex fan of the acceleration card and / or the system fan of the server according to the relationship between the surface temperature of each acceleration card of the server and the first temperature threshold T1 and the second temperature threshold T2; wherein, the first temperature threshold T1 is less than the second temperature threshold T2.

[0068] Specifically, according to the relationship between the surface temperature of the acceleration card and the two hierarchical thresholds T1 and T2 (T1 < T2), the GPU firmware preferentially starts its own built-in vortex fan at T1, and only when the temperature continues to rise to T2, the system fan is started by the controller (BMC), achieving the purpose of time-sequential hierarchical response of GPU local heat dissipation and system-level heat dissipation. Thereby, the system fan remains at a low speed or is turned off during the low and medium load stages, significantly reducing the power consumption and noise of the whole machine, and solving the technical problems of redundant operation, high virtual energy consumption, and increased mechanical wear caused by the synchronous start and stop of the system fan and the vortex fan of the acceleration card in existing servers. At the same time, by introducing a clear dual-threshold mechanism (T1 < T2) in the temperature control of the acceleration card and defining the response order of the control subject, the purpose of eliminating the control conflict and limit cycle oscillation caused by the overlapping temperature response of the vortex fan and the system fan of the acceleration card is achieved, thereby ensuring that the two-stage fan system operates smoothly during dynamic load changes and the speed switching is oscillation-free, reducing the temperature fluctuation amplitude.

[0069] In this embodiment, the method for controlling the start of the turbofan and / or system fan of the acceleration card according to the relationship between the surface temperature of each acceleration card and the first temperature threshold T1 and the second temperature threshold T2 includes:

[0070] If the surface temperature of each acceleration card continues to rise and reaches the first temperature threshold T1, start the turbofan of the acceleration card; if the surface temperature of each acceleration card continues to rise and reaches the second temperature threshold T2, then start the system fan.

[0071] Specifically, in the temperature control of the acceleration card, a step-by-step response mechanism of "starting the local turbofan first and then the system fan" is introduced, and T1 and T2 are used as two-level trigger conditions (T1 < T2) respectively under the condition of continuous temperature rise, achieving the purpose that the PU heat dissipation load is preferentially borne by the local efficient turbofan and the system fan only intervenes as a redundant backup. Moreover, in medium and low load conditions, the system fan remains stopped or runs at a low speed, significantly reducing the machine power consumption and noise. At the same time, by setting the continuous temperature rise as the prerequisite for starting the system fan and constructing a control buffer with a clear threshold difference of T2 > T1, it is avoided that the system fan is mis-started due to instantaneous temperature fluctuations or sensor jitters, thus realizing the stable operation of the two-level fan system in the dynamic change of the load, without repeated start-stop and rotational speed oscillation, and significantly improving the temperature control stability.

[0072] In this embodiment, the method for controlling the start of the turbofan and / or system fan of the acceleration card according to the relationship between the surface temperature of each acceleration card and the first temperature threshold T1 and the second temperature threshold T2 further includes:

[0073] If the surface temperature of each acceleration card continues to drop and drops below the second temperature threshold T2, reduce the rotational speed of the system fan; if the surface temperature of each acceleration card continues to drop and drops below the first temperature threshold T1, turn off the system fan and / or reduce the rotational speed of the turbofan of the acceleration card.

[0074] Specifically, in the temperature drop stage, a reverse hierarchical control logic of "reducing the system fan first and then turning off / reducing the turbofan" is set (T2 is the system fan speed reduction point, and T1 is the turbofan turn-off point), achieving the purpose of decoupling the heat dissipation system from the load change synchronously and avoiding the sudden stop or over-response of the fan caused by instantaneous temperature fluctuations, ensuring that the system fan quickly reduces its speed after the GPU load drops and the turbofan is smoothly turned off after the waste heat dissipates, significantly reducing the ineffective energy consumption and mechanical impact. At the same time, by corresponding the turn-off / speed reduction actions of the system fan and the turbofan to the "continuous temperature drop" state and the sequential triggering of the double thresholds from T2 to T1, the purpose of eliminating the interference of temperature feedback lag and sensor noise on the control decision is achieved, thus ensuring that the two-level fan system realizes a smooth, progressive and non-oscillating collaborative exit during the load change, and further maintaining the stable junction temperature of the acceleration card within the safe range.

[0075] In this embodiment, the accelerator card temperature control method includes reducing the speed of the system fan or the turbofan by no more than 5 to 10% of the rated speed per second.

[0076] Specifically, by limiting the rate at which the speed of the system fan or the turbofan of the accelerator card is reduced to no more than 5% to 10% of the rated speed per second, the accelerator card's (GPU card) junction temperature is prevented from dropping sharply due to a sudden drop in speed, thus achieving a gradual matching of heat dissipation power during load reduction and maintaining a smooth transition of GPU junction temperature without overshoot fluctuations.

[0077] In this embodiment, during the process of controlling the temperature of all accelerator cards, the accelerator card temperature control method of all accelerator cards is executed synchronously.

[0078] Specifically, during the temperature control process of all accelerator cards, the temperature control methods of all GPUs (including turbofan start / stop, system fan linkage, speed adjustment, etc.) are set to be executed synchronously, thereby eliminating the uneven thermal load between GPUs caused by control cycle misalignment, polling delay or firmware response differences. This ensures that all GPUs operate uniformly at the same temperature threshold trigger point, avoiding some GPUs from overheating due to control lag and some GPUs from overcooling due to control overshoot.

[0079] In this embodiment, the heat dissipation control method further includes a temperature monitoring method for a multi-accelerator card server serial polling architecture, which includes:

[0080] Within the current polling period K, obtain the maximum temperature value T of all accelerator cards. MAX (K), and record the average temperature T of all accelerator cards in the first two polling cycles. AVG (K-1) and T AVG (K-2); and according to the fifth formula:

[0081] FS(k) = FS act (k-1)+K p (T MAX (K)-T AVG (K-1))+K i (T MAX (K)-T sp )+K d (T MAX (K)-2 T AVG (K-1)+T AVG(K-2) Obtain the pulse width modulation output value FS(k) of the server's system fan within the current polling period K, and control the system fan speed based on the pulse width modulation output value FS(k); where FS act (k-1) represents the pulse width modulation output value of the server's system fan in the previous polling cycle, T sp To set the temperature threshold, K p The first coefficient is 1 ≤ K p ≤20, K i It is the second coefficient and satisfies 0.02≤K i ≤0.50, K d It is the third coefficient and satisfies 0.1≤K d ≤4.0.

[0082] Specifically, within each polling cycle K, the maximum temperature T of all accelerator cards (GPU cards) is extracted. MAX (K), and combined with the average temperature T of all accelerator cards in the first two polling cycles. AVG (K-1) and T AVG Using (K-2) as the input parameter, the global thermal peak value is used instead of a single delayed temperature point for control decisions, thereby eliminating the temperature reading lag effect caused by serial polling. This achieves a millisecond-level acceleration of the system fan's response to the GPU's instantaneous thermal shock, thus solving the problem of temperature detection delay and thermal response lag caused by temperature monitoring systems using serial polling architectures in related technologies. Simultaneously, by using T... MAX (K) serves as the dynamic error benchmark, replacing the static target temperature T. sp The instantaneous deviation, and construct a model based on the temperature change trend (T) MAX (K)-T AVG The differential feedback mechanism (K-1) suppresses control oscillations and over-adjustment caused by temperature fluctuations, thereby achieving a significant improvement in the smoothness and stability of the system fan speed regulation.

[0083] In this embodiment, based on the maximum temperature value T of all accelerator cards MAX (K) and the average temperature T of all accelerator cards in the previous polling cycle AVG The relationship between (K-1) adjusts the pulse width modulation output value FS(k) of the system fan.

[0084] Specifically, the highest temperature T of all GPU cards within the current polling period K. MAX (K) is used as a real-time thermal shock indicator, and is compared with the average temperature T of all GPU cards in the previous polling cycle. AVG(K-1) performs difference comparison to avoid single-point temperature delay caused by serial polling, directly capturing the transient changes of the hottest GPU in the system. This achieves millisecond-level perception and early warning of sudden power surges in the GPU, thus enabling the system fan to actively accelerate before the temperature reaches its peak. Simultaneously, using T... MAX (K)-T AVG (K-1) is used as the core error input to replace the static deviation between the traditional single-point temperature and the set threshold, thus constructing a "thermal trend differential feedback" mechanism. This allows the system fan response to focus on the rate of temperature change rather than the absolute value, thereby suppressing false triggering and over-adjustment caused by asynchronous temperature sampling or uneven load distribution. This achieves the technical effect of simultaneously improving the dynamic adaptability and stability of the system fan speed regulation.

[0085] In this embodiment, based on the maximum temperature value T of all accelerator cards MAX (K) and the average temperature T of all accelerator cards in the previous polling cycle AVG The methods for adjusting the pulse width modulation output value FS(k) of the system fan based on the relationship between (K-1) include:

[0086] If the maximum temperature value T of all accelerator cards MAX (K) and the average temperature T of all accelerator cards in the previous polling cycle AVG If there is a fixed deviation A between (K-1), then the pulse width modulation output value FS'(k) of the system fan will be adjusted to: FS'(k) = FS(k) + K p A.

[0087] Specifically, based on the system fan pulse width modulation output value FS(k), a fixed deviation correction term K is calculated based on historical data statistics. p A fine-tunes the pulse width modulation output value of the system fan, achieving the goal of compensating for systemic temperature measurement offsets caused by uneven distribution of GPU temperature sensors, differences in heat dissipation paths, or slight asymmetry in load distribution. This significantly improves the steady-state accuracy and consistency of temperature feedback without altering the main control law, thereby solving the technical problem in traditional servers where a fixed deviation (e.g., 3-5℃) between the maximum and average values ​​leads to delayed fan response or premature speed reduction, and aggravated temperature fluctuations in critical components. Simultaneously, by proportionally increasing the fixed deviation A with a gain K... p By superimposing the signal onto the main output, dynamic calibration of the temperature gradient trend is achieved, thus eliminating steady-state temperature errors caused by sensor drift or airflow asymmetry during long-term operation while maintaining a rapid response to thermal shock.

[0088] In this embodiment, based on the maximum temperature value T of all accelerator cardsMAX (K) and the average temperature T of all accelerator cards in the previous polling cycle AVG The methods for adjusting the pulse width modulation output value FS(k) of the system fan based on the relationship between (K-1) include:

[0089] If the maximum temperature value T of all accelerator cards MAX (K) and the average temperature T of all accelerator cards in the previous polling cycle AVG If the values ​​(K-1) are proportional, then the pulse width modulation output value FS'(k) of the system fan will be adjusted as follows:

[0090] FS'(k) = FS(k) + K p T MAX (K) (B-1); where B is a proportionality coefficient greater than 1.

[0091] Specifically, by introducing a value based on the maximum temperature T into the system fan pulse width modulation output value FS(k). MAX (K) proportional amplification term K p T MAX (K) (B-1) achieves the goal of nonlinearly enhancing the response to thermal risks under high-temperature conditions. This results in a proportionally accelerated system fan speed during periods of rapid GPU load increase and junction temperature rise, thus solving the technical problems of traditional control methods that suffer from sluggish response and insufficient airflow under high thermal loads, leading to GPU overheating and protective frequency reduction or shutdown. Furthermore, by setting the proportional coefficient B to a constant greater than 1 and relating it to T... MAX (K) is multiplied and then superimposed on the main control output, realizing dynamic gain compensation for the heat peak under non-uniform temperature distribution. This achieves the goal of significantly improving the heat dissipation margin during high load while maintaining stable control during low load. Thus, the system fan control achieves the technical effect of "intelligent gain adjustment" across the entire operating range.

[0092] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for implementing the heat dissipation control method for a server as described above when executing the computer program.

[0093] As can be seen from the above description, the embodiments of the present invention achieve the following technical effects:

[0094] When the server is operating, the system fan exhausts hot air through the rear window of the chassis, creating negative pressure, while the front window draws in cool air. A first air duct, located at the front window, directly guides the airflow to the battery area of ​​one or more network cards, ensuring critical components receive sufficient cooling. Simultaneously, a second air duct is positioned above all accelerator cards, blocking the gaps between adjacent cards and effectively preventing bypasses of cool air leakage. This forces airflow through the accelerator cards' built-in turbofans for efficient cooling. Through this optimized airflow organization, the system fan does not need to maintain excessively high speeds to compensate for air leakage. This significantly reduces ineffective energy consumption of the system fan while ensuring the temperature stability of internal components such as accelerator cards and network cards. It solves the problems of high system fan energy consumption and poor temperature stability of internal components in GPU servers, improving the temperature stability of internal components, avoiding frequent acceleration and deceleration of the system fan, and extending the system fan's lifespan.

[0095] Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this application. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A server, characterized in that, include: Multiple accelerator cards (20), each of the accelerator cards (20) including a turbofan; One or more network interface cards (30); A first air guide (40) is located above the one or more network cards (30) to guide the incoming airflow to the battery area (31) of the one or more network cards (30). A second air guide shroud (50) is disposed above at least two of the accelerator cards (20) and blocks the gap between at least two adjacent accelerator cards (20) to guide airflow to the turbofan after the system fan is started.

2. The server according to claim 1, characterized in that, The first air guide hood (40) includes a first plate (41), an inclined guide plate (42) and a second plate (43) connected in sequence. The first plate (41) is located at the air inlet, and the inclined guide plate (42) is set at an obtuse angle to the first plate (41) and the second plate (43).

3. The server according to claim 2, characterized in that, The first air guide shroud (40) is made of at least one of metal, ABS, PP and silicone foam; or the inclined air guide plate (42) is made of silicone foam or rubber.

4. The server according to claim 2, characterized in that, The server also includes a chassis (10), the first air guide shroud (40) is located inside the chassis (10), the chassis (10) includes multiple columns and a crossbeam set on the columns; the first air guide shroud (40) also includes a bending plate (44), the bending plate (44) is set at the end of the first plate (41) away from the inclined guide plate (42) and is set at an angle with the first plate (41); wherein the bending plate (44) and the first plate (41) surround to form an overlapping part, the overlapping part overlaps on the crossbeam.

5. The server according to claim 4, characterized in that, The elastic modulus E of the material selected for the first air guide shroud (40) is obtained from the first formula: I req =I old +[ΔK bending W L 2 ] / (α HAVE BEEN); wherein I req is the cross-sectional moment of inertia of the rear single column needed to compensate for old is the cross-sectional moment of inertia of the front single column, W is the width of the cabinet (10), L is the internal height of the cabinet (10), a is the support coefficient of the first air deflector (40) and satisfies: 0.1≤a≤1.0, ΔK bending is the bending stiffness deficiency and is derived from the second formula: ΔK bending =(F / δ max )-(F / δ actual F is the total load of the chassis (10), δ max For the maximum allowable deflection, δ actual This represents the actual deflection. If the elastic modulus E of the material selected by the first air guide (40) is less than the first threshold or within the first range, then the first air guide (40) selects the first material; if the elastic modulus E of the material selected by the first air guide (40) is greater than or equal to the first threshold and less than or equal to the second threshold or within the second range, then the first air guide (40) selects the second material.

6. The server according to claim 1, characterized in that, The server also includes a chassis (10), the first air guide shroud (40) is located inside the chassis (10); a plurality of acceleration cards (20) are spaced apart along a preset direction and adjacent acceleration cards (20) are parallel to each other; the preset direction is set at an angle to the air intake direction of the chassis (10).

7. The server according to claim 6, characterized in that, The second air guide shroud (50) includes: Cover plate (51); Multiple insert plates (52) are spaced apart along the length of the cover plate (51) on the lower surface of the cover plate (51), and each insert plate (52) extends between two adjacent accelerator cards (20).

8. The server according to claim 7, characterized in that, The second air guide shroud (50) also includes: A support column (53) is disposed on the lower surface of the cover plate (51) for supporting the cover plate (51).

9. The server according to claim 7, characterized in that, The server also includes a chassis (10), and the first air duct (40) is located inside the chassis (10); the cover plate (51) includes: The third plate (511) has a plurality of insert plates (52) disposed on the lower plate surface of the third plate (511); A partition (512) is disposed on the upper surface of the third plate (511), and the partition (512) contacts the side wall of the chassis (10) to support the side wall.

10. The server according to claim 9, characterized in that, The second air guide cover (50) is made of at least one of metal, ABS, PP and silicone foam; and / or, the second air guide cover (50) is a one-piece molded structure.

11. The server according to claim 10, characterized in that, The elastic modulus E' of the material selected for the second air guide shroud (50) satisfies the third formula: E'≥(ΔK W 3 ) / (β t 3 H); Where W is the width of the chassis (10), β is the support coefficient of the second air guide shroud (50) and satisfies: 0.8≤β≤1.2, t is the thickness of the enclosure (512), H is the height of the chassis (10), and ΔK is the strength defect value to be compensated and is obtained from the fourth formula: ΔK=(q S) (1 / δ max -1 / δ actual ); q is the uniformly distributed load acting on the side wall of the chassis (10), S is the effective force-bearing area of ​​the side wall of the chassis (10), δ max δ is the maximum permissible safe deflection at the center of the side wall of the chassis (10). actual The actual deflection of the side wall of the chassis (10) under load q when it is not reinforced.

12. A heat dissipation control method for a server, characterized in that, Applied to the server according to any one of claims 1 to 11, the heat dissipation control method includes a two-stage fan control method, the two-stage fan control method comprising: The turbo fan of the accelerator card (20) and / or the system fan of the server are started according to the relationship between the surface temperature of each accelerator card (20) of the server and the first temperature threshold T1 and the second temperature threshold T2. Wherein, the first temperature threshold T1 is less than the second temperature threshold T2.

13. The heat dissipation control method according to claim 12, characterized in that, The method for controlling the start-up of the turbofan and / or the system fan of the accelerator card (20) based on the relationship between the surface temperature of each accelerator card and the first temperature threshold T1 and the second temperature threshold T2 includes: If the surface temperature of each of the accelerator cards (20) continues to rise and reaches the first temperature threshold T1, the turbofan of the accelerator card (20) is started; if the surface temperature of each of the accelerator cards (20) continues to rise and reaches the second temperature threshold T2, the system fan is started again.

14. The heat dissipation control method according to claim 12, characterized in that, The method for controlling the start-up of the turbofan and / or the system fan of the accelerator card (20) based on the relationship between the surface temperature of each accelerator card (20) and the first temperature threshold T1 and the second temperature threshold T2 further includes: If the surface temperature of each of the accelerator cards (20) continues to drop and falls below the second temperature threshold T2, the speed of the system fan is reduced; if the surface temperature of each of the accelerator cards (20) continues to drop and falls below the first temperature threshold T1, the system fan is turned off and / or the speed of the turbofan of the accelerator card (20) is reduced.

15. The heat dissipation control method according to claim 14, characterized in that, Methods for reducing the speed of the system fan or turbofan include: reducing the speed of the system fan or turbofan by no more than 5 to 10% of the rated speed per second.

16. The heat dissipation control method according to claim 12, characterized in that, During the process of controlling the temperature of all accelerator cards, the temperature control methods for all accelerator cards are executed synchronously.

17. The heat dissipation control method according to claim 12, characterized in that, The heat dissipation control method further includes a temperature monitoring method for a multi-accelerator card server serial polling architecture, which includes: Within the current polling period K, obtain the maximum temperature value T of all accelerator cards. MAX (K), and record the average temperature T of all accelerator cards in the first two polling cycles. AVG (K-1) and T AVG (K-2); and according to the fifth formula: FS(k) = FS act (k-1)+K p (T MAX (K)-T AVG (K-1))+K i (T MAX (K)-T sp )+K d (T MAX (K)-2 T AVG (K-1)+T AVG (K-2) Obtain the pulse width modulation output value FS(k) of the system fan of the server in the current polling period K, and control the speed of the system fan based on the pulse width modulation output value FS(k); Among them, FS act (k-1) represents the pulse width modulation output value of the server's system fan in the previous polling cycle, T sp To set the temperature threshold, K p The first coefficient is 1 ≤ K p ≤20, K i It is the second coefficient and satisfies 0.02≤K i ≤0.50, K d It is the third coefficient and satisfies 0.1≤K d ≤4.

0.

18. The heat dissipation control method according to claim 17, characterized in that, Based on the maximum temperature value T of all accelerator cards MAX (K) and the average temperature T of all accelerator cards in the previous polling cycle AVG The relationship between (K-1) adjusts the pulse width modulation output value FS(k) of the system fan.

19. The heat dissipation control method according to claim 18, characterized in that, Based on the maximum temperature value T of all accelerator cards MAX (K) and the average temperature T of all accelerator cards in the previous polling cycle AVG The methods for adjusting the pulse width modulation output value FS(k) of the system fan based on the relationship between (K-1) include: If the maximum temperature value T of all accelerator cards MAX (K) and the average temperature T of all accelerator cards in the previous polling cycle AVG If there is a fixed deviation A between (K-1), then the pulse width modulation output value FS'(k) of the system fan will be adjusted as follows: FS’(k)=FS(k)+K p A。 20. The heat dissipation control method according to claim 18, characterized in that, Based on the maximum temperature value T of all accelerator cards MAX (K) and the average temperature T of all accelerator cards in the previous polling cycle AVG The methods for adjusting the pulse width modulation output value FS(k) of the system fan based on the relationship between (K-1) include: If the maximum temperature value T of all accelerator cards MAX (K) and the average temperature T of all accelerator cards in the previous polling cycle AVG If the values ​​(K-1) are proportional, then the pulse width modulation output value FS'(k) of the system fan will be adjusted as follows: FS'(k) = FS(k) + K p T MAX (K) (B-1); where B is a proportionality coefficient greater than 1.

21. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the heat dissipation control method for a server as described in any one of claims 12 to 20 when executing the computer program.

Images