Cold plate type liquid cooling module for intelligent calculation center

Abstract

The invention discloses a cold plate type liquid cooling module for an intelligent calculation center, and relates to the technical field of liquid cooling, the cold plate type liquid cooling module comprises a water pump and a liquid cooling shell, the outlet end of the water pump is in through connection with a main control valve, the outlet end of the main control valve is in through connection with a plurality of two-position two-way reversing valves, the first port of each two-position two-way reversing valve is connected with the main control valve, and the second port of each two-position two-way reversing valve is connected with the liquid cooling shell. A second port of the two-position two-way reversing valve is connected with the liquid cooling shell in a penetrating mode, one end of the liquid cooling shell is connected with a third port of the two-position two-way reversing valve in a penetrating mode, a fourth port of the two-position two-way reversing valve is connected with a refrigerator in a penetrating mode, one end of the refrigerator is connected with a water pump in a penetrating mode, and the liquid cooling shell is installed on a GPU operation chip in an attached mode. A flow meter is connected between a fourth port of the two-position two-way reversing valve and the refrigerator in a penetrating mode, and the two ends of the liquid cooling shell are connected with a throttling valve and a flow divider valve in a penetrating mode correspondingly.

Description

Technical Field

[0001] The present invention relates to the field of liquid cooling technology, and specifically to a cold plate liquid cooling module for an intelligent computing center. Background Art

[0002] The cold plate liquid cooling module of the intelligent computing center is an efficient heat dissipation solution, which is particularly suitable for high-power density computing equipment, such as large data centers, cloud computing platforms and artificial intelligence training scenarios. The liquid cooling module attached to the GPU of the computing power server can reduce heat dissipation noise.

[0003] When using the existing liquid cooling module, the direction of the liquid is fixed, and the temperature of the coolant at the inlet will be lower than the temperature at the outlet. Long-term cooling will cause uneven heat distribution on the bonding surface of the GPU computing chip and the liquid cooling module. Due to the thermal expansion and contraction effect, the GPU computing chip will accumulate tiny deformations, which will reduce its lifespan. The existing solution is to change the direction of the coolant pumping, which will cause the water pump to stop at regular intervals and discontinuous heat dissipation. Therefore, it is necessary to design a cold plate liquid cooling module for the intelligent computing center with uniform heat dissipation. Summary of the invention

[0004] The purpose of the present invention is to provide a cold plate liquid cooling module for an intelligent computing center to solve the problems raised in the above-mentioned background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a cold plate liquid cooling module for an intelligent computing center, comprising a water pump and a liquid cooling shell, the outlet end of the water pump is through-connected with a main control valve, the outlet end of the main control valve is through-connected with a plurality of two-position two-way reversing valves, the first port of the two-position two-way reversing valve is connected to the main control valve, the second port of the two-position two-way reversing valve is through-connected with the liquid cooling shell, one end of the liquid cooling shell is through-connected with the third port of the two-position two-way reversing valve, the fourth port of the two-position two-way reversing valve is through-connected with a refrigerator, one end of the refrigerator is through-connected with the water pump, and the liquid cooling shell is fittedly installed on the GPU computing chip.

[0006] According to the above technical solution, a flow meter is connected between the fourth port of the two-position two-way reversing valve and the refrigerator, and a throttle valve and a diverter valve are respectively connected at both ends of the liquid cooling shell. One end of the diverter valve is connected to the refrigerator, and one end of the throttle valve is connected to the second port of the two-position two-way reversing valve.

[0007] . According to the cold plate liquid cooling module for the intelligent computing center, the feature is that: the control system of the liquid cooling module includes a flow calculation module, a shunt control module, a valve reversing module, a GPU state monitoring module, and a heat generation estimation module, the flow calculation module is electrically connected to the throttle valve, the shunt control module is electrically connected to the shunt valve, the valve reversing module is electrically connected to the two-position two-way reversing valve, and the GPU state monitoring module is electrically connected to the heat generation estimation module and the flow calculation module;

[0008] The flow calculation module is used to calculate the coolant flow passing through each liquid cooling shell, the diversion control module is used to control the diversion valve to distribute the flow of coolant flowing out of the liquid cooling shell, the valve reversing module is used to control each two-position two-way reversing valve to perform reversal, the GPU status monitoring module is used to monitor the working status of the GPU computing power chip, and the heat generation estimation module is used to estimate the heat generation according to the working status of the GPU computing power chip, thereby deriving the coolant flow that needs to be pumped to the liquid cooling shell.

[0009] According to the above technical solution, the working method of the control system of the liquid cooling module is:

[0010] S1, start the water pump to mobilize the coolant flow, at this time the main control valve is open, at this time the coolant enters from the first port of the two-position two-way reversing valve, flows to the liquid cooling shell from the second port, flows out of the liquid cooling shell and enters the third port, and is discharged from the fourth port to the refrigerator to pump the coolant forward;

[0011] S2. When reversing, the valve reversing module controls each two-position two-way reversing valve to switch the flow direction of the coolant. At this time, the coolant enters from the first port of the two-position two-way reversing valve, is discharged from the third port and flows to the liquid cooling shell, and then enters the second port after being discharged from the liquid cooling shell and enters the two-position two-way reversing valve, and is discharged from the fourth port to the refrigerator, and the coolant is pumped in reverse to achieve uniform flow of the coolant in the liquid cooling shell;

[0012] S3. Obtain the real-time working status of the GPU computing chip through the GPU status monitoring module, monitor the computing volume and GPU utilization rate of the GPU computing chip in artificial intelligence training, calculate the coolant flow required by each liquid cooling shell matching each GPU computing chip, and realize flow control through the throttle valve;

[0013] S4. When pumping coolant in the forward direction, the opening of each diverter valve is adjusted so that excess coolant can flow back to the refrigerator through the diverter valve, so that the pumping flow of the water pump remains unchanged, avoiding the problem of repeated adjustment of the pumping volume due to uneven outlet pressure. When pumping coolant in the reverse direction, the throttle valve opening is maximized, and the coolant is quickly pumped in the reverse direction under the condition of maximum flow to neutralize the uneven temperature. After a period of time, it returns to the forward pumping state.

[0014] According to the above technical solution, in S3, the specific calculation method of the coolant flow rate is:

[0015] S3-1. Enable the API tool in the GPU status monitoring module to obtain the working status data of each GPU in real time. First, record the current computing amount being processed, obtain it through the number of floating-point operations k processed per second, and monitor the GPU utilization rate at the same time to obtain the real-time GPU utilization rate n%. Based on these two indicators, the heat generation is calculated using the formula Q=αk+lgβ(n%+1), where α is the weight coefficient of the number of floating-point operations, and β is the weight coefficient of the GPU utilization rate;

[0016] S3-2. According to the estimated calorific value and the specific heat capacity of the coolant, the required coolant flow rate is calculated. The formula is coolant flow rate L = μQ, where μ is the specific heat capacity coefficient of the coolant. The coolant flow rate L of each liquid-cooled shell 6 is calculated independently to obtain the coolant flow rate set {L1, L2, ..., L x}, where x is the number of liquid cooling shells.

[0017] According to the above technical solution, in S4, the specific method for adjusting the opening of each diverter valve is: first determine the pumping flow rate L0 of the water pump, and then adjust the opening of each diverter valve according to the coolant flow rate set {L1, L2, ..., L x}Calculate the actual total coolant flow ∑L x , then calculate the average flow rate of each liquid cooling shell The flow rate divided by each diverter valve is where i∈[1~x], L i is the coolant flow rate of the liquid-cooled shell with serial number i.

[0018] According to the above technical solution, the control system of the liquid cooling module also includes a heat impact correction module, which is used to correct the coolant flow of the liquid cooling shell according to the heat impact of the GPU computing chip in the adjacent position. The heat impact correction module includes a coordinate recording module, a working time recording module, an adjacent coordinate judgment module, and a heat accumulation calculation module. The coordinate recording module is used to record the coordinate position of each liquid cooling shell, the working time recording module is used to record the working time of each GPU computing chip, the adjacent coordinate judgment module is used to judge the number of GPU computing chips adjacent to each GPU computing chip, and the heat accumulation calculation module is used to calculate the cumulative impact of heat based on the number of adjacent computing chips and working time.

[0019] According to the above technical solution, the control system of the liquid cooling module also includes a heat impact correction module, which is used to correct the coolant flow of the liquid cooling shell according to the heat impact of the GPU computing chip in the adjacent position. The heat impact correction module includes a coordinate recording module, a working time recording module, an adjacent coordinate judgment module, and a heat accumulation calculation module. The coordinate recording module is used to record the coordinate position of each liquid cooling shell, the working time recording module is used to record the working time of each GPU computing chip, the adjacent coordinate judgment module is used to judge the number of GPU computing chips adjacent to each GPU computing chip, and the heat accumulation calculation module is used to calculate the cumulative impact of heat based on the number of adjacent computing chips and working time.

[0020] According to the above technical solution, the working method of the heating effect correction module is:

[0021] S5-1. When recording the coordinates of the adjacent chip positions of a certain chip position, let the horizontal coordinate parameter of the chip position be X, and the vertical coordinate parameter be Y. The current chip position is located in the ath column from left to right and the bth row from bottom to top of a mounting rack on a certain side of the intelligent computing center. The coordinates of the current chip position are X. a ,Y b , the chip position coordinates adjacent to it are X a+1 ,Y b , X a ,Y b+1 , X a―1 ,Y b , X a ,Y b―1 , the adjacent chip positions located at the peripheral chip positions need to subtract the non-existent coordinates;

[0022] S5-2. When the GPU computing chip at the adjacent position of the current chip is working, it will emit heat and cause a temperature rise effect Q0=εt on the current GPU computing chip, where ε is the heat accumulation coefficient caused by the continuous operation of the GPU computing chip at the adjacent position, and t is the working time of the GPU computing chip at the adjacent position. The temperature rise effect caused by all the surrounding GPU computing chips is Where j is the number of GPU computing chips adjacent to a GPU computing chip, so the corrected The adjacent temperature rise is taken into account in the coolant pumping flow rate through calculation, so that the coolant pumping can more accurately match the current working environment.

[0023] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: the present invention connects two-position two-way reversing valves at both ends of the liquid cooling shell in the liquid cooling module, and switches the connection mode of the two-position two-way reversing valve, so that when the water pump continues to pump, the liquid cooling shell only changes the flow direction of the coolant flowing through the liquid cooling shell, thereby achieving uniform flow of the coolant, and does not cause the phenomenon that the inlet temperature is always lower than the outlet temperature, thereby avoiding the accumulation of thermal expansion and contraction effects. In addition, by estimating the heat generation according to the working state of the GPU computing power chip, the coolant flow rate of each liquid cooling shell is controlled separately, so that the cooling effect is better, and the coolant flow rate of each liquid cooling shell is corrected by the degree of mutual influence of adjacent GPU computing power chips due to heat diffusion, so as to offset the effect of the superimposed temperature rise of adjacent computing power chips. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0025] Figure 1 It is a schematic diagram of the overall hydraulic principle of the present invention;

[0026] Figure 2 It is the installation schematic diagram of the present invention;

[0027] Figure 3 It is a system module diagram of the present invention;

[0028] 1. Main control valve; 2. Water pump; 3. Refrigerator; 4. Two-position two-way reversing valve; 5. Throttle valve; 6. Liquid cooling shell; 7. Diverter valve; 8. Flow meter. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0030] See also Figure 1-Figure 3The present invention provides a technical solution: a cold plate liquid cooling module for an intelligent computing center, comprising a water pump 2 and a liquid cooling shell 6, wherein the outlet end of the water pump 2 is connected with a main control valve 1, and the outlet end of the main control valve 1 is connected with a plurality of two-position two-way reversing valves 4, wherein the first port of the two-position two-way reversing valve 4 is connected with the main control valve 1, and the second port of the two-position two-way reversing valve 4 is connected with the liquid cooling shell 6, and one end of the liquid cooling shell 6 is connected with the third port of the two-position two-way reversing valve 4, and the fourth port of the two-position two-way reversing valve 4 is connected with a refrigerator 3, and one end of the refrigerator 3 is connected with the water pump 2, and the liquid cooling shell 6 is fitted on the GPU computing chip to dissipate heat from the GPU computing chip using the cooling liquid flowing inside;

[0031] A flow meter 8 is connected between the fourth port of the two-position two-way reversing valve 4 and the refrigerator 3, and a throttle valve 5 and a diverter valve 7 are connected at both ends of the liquid cooling shell 6, one end of the diverter valve 7 is connected to the refrigerator 3, and one end of the throttle valve 5 is connected to the second port of the two-position two-way reversing valve 4;

[0032] The control system of the liquid cooling module includes a flow calculation module, a shunt control module, a valve reversing module, a GPU state monitoring module, and a calorific value estimation module. The flow calculation module is electrically connected to the throttle valve 5, the shunt control module is electrically connected to the shunt valve 7, the valve reversing module is electrically connected to the two-position two-way reversing valve 4, and the GPU state monitoring module is electrically connected to the calorific value estimation module and the flow calculation module.

[0033] The flow calculation module is used to calculate the coolant flow through each liquid cooling shell 6, the flow distribution control module is used to control the flow distribution valve 7 to distribute the flow of coolant flowing out of the liquid cooling shell 6, the valve reversing module is used to control each two-position two-way reversing valve 4 to perform reversal, the GPU status monitoring module is used to monitor the working status of the GPU computing chip, and the heat generation estimation module is used to estimate the heat generation according to the working status of the GPU computing chip, thereby obtaining the coolant flow that needs to be pumped to the liquid cooling shell 6;

[0034] The working method of the control system of the liquid cooling module is:

[0035] S1, start the water pump 2 to mobilize the flow of coolant, at this time the main control valve 1 is opened, at this time the coolant enters from the first port of the two-position two-way reversing valve 4, flows from the second port to the liquid cooling shell 6, flows out from the liquid cooling shell 6 into the third port, and is discharged from the fourth port to the refrigerator 3 to pump the coolant forward;

[0036] S2, when reversing, the valve reversing module controls each two-position two-way reversing valve 4 to switch the flow direction of the coolant. At this time, the coolant enters from the first port of the two-position two-way reversing valve 4, is discharged from the third port and flows to the liquid cooling shell 6, and then enters the second port after being discharged from the liquid cooling shell 6 and enters the two-position two-way reversing valve 4, and is discharged from the fourth port to the refrigerator 3, and the coolant is pumped in reverse to achieve uniform flow of the coolant in the liquid cooling shell 6;

[0037] S3, obtaining the real-time working status of the GPU computing chip through the GPU status monitoring module, monitoring the computing volume and GPU utilization rate of the GPU computing chip in artificial intelligence training, calculating the cooling liquid flow required by each liquid cooling shell 6 matching each GPU computing chip, and realizing flow control through the throttle valve 5;

[0038] S4. When the coolant is pumped forward, the opening of each diverter valve 7 is adjusted so that the excess coolant flows back to the refrigerator 3 through the diverter valve 7, so that the pumping flow of the water pump 2 remains unchanged, avoiding the problem that the pumping amount of the water pump 2 needs to be repeatedly adjusted due to uneven outlet pressure. When the coolant is pumped backward, the throttle valve 5 is opened to the maximum, and the coolant is quickly pumped backward under the condition of maximum flow to neutralize the uneven temperature. After a period of time, the forward pumping state is returned;

[0039] In S3, the specific calculation method of the coolant flow rate is:

[0040] S3-1. Enable the API tool in the GPU status monitoring module to obtain the working status data of each GPU in real time. First, record the current computing amount being processed, obtain it through the number of floating-point operations k processed per second, and monitor the GPU utilization rate at the same time to obtain the real-time GPU utilization rate n%. Based on these two indicators, the heat generation is calculated using the formula Q=αk+lgβ(n%+1), where α is the weight coefficient of the number of floating-point operations, and β is the weight coefficient of the GPU utilization rate;

[0041] S3-2. According to the estimated calorific value and the specific heat capacity of the coolant, the required coolant flow rate is calculated. The formula is coolant flow rate L = μQ, where μ is the specific heat capacity coefficient of the coolant. The coolant flow rate L of each liquid-cooled shell 6 is calculated independently to obtain the coolant flow rate set {L1, L2, ..., L x}, where x is the number of liquid cooling shells 6; direct temperature measurement may have a link lag, especially when the load changes rapidly, the temperature rise may lag behind the actual heat generation, therefore, relying on temperature measurement may cause the cooling response to be slower than the actual need, this step can adjust the coolant flow before the temperature rises, thereby avoiding excessive temperature, because it is adjusted directly by controlling the opening of the throttle valve 5 connected to the liquid cooling shell 6, the adjustment is highly real-time and the response speed is accelerated;

[0042] In S4, the specific method for adjusting the opening of each diverter valve 7 is: first determine the pumping flow rate L0 of the water pump 2, and then adjust the opening of each diverter valve 7 according to the coolant flow rate set {L1, L2, ..., L x}Calculate the actual total coolant flow ∑L x , then calculate the average flow rate of each liquid cooling shell 6 The flow rate divided by each diverter valve 7 is where i∈[1~x], L i is the coolant flow rate of the liquid cooling shell 6 with serial number i; by accurately controlling the flow rate of the diverter valve 7, the pumping flow rate of the water pump 2 is always kept consistent, the water pump 2 works more smoothly, and the working stability of the entire hydraulic system is improved;

[0043] The control system of the liquid cooling module also includes a heat impact correction module, which is used to correct the coolant flow of the liquid cooling shell 6 according to the heat impact of the GPU computing chip at the adjacent position. The heat impact correction module includes a coordinate recording module, a working time recording module, an adjacent coordinate judgment module, and a heat accumulation calculation module. The coordinate recording module is used to record the coordinate position of each liquid cooling shell 6, the working time recording module is used to record the working time of each GPU computing chip, the adjacent coordinate judgment module is used to judge the number of GPU computing chips adjacent to each GPU computing chip, and the heat accumulation calculation module is used to calculate the cumulative impact of heat according to the number of adjacent computing chips and working time;

[0044] The working method of the heating effect correction module is:

[0045] S5-1. When recording the coordinates of the adjacent chip positions of a certain chip position, let the horizontal coordinate parameter of the chip position be X, and the vertical coordinate parameter be Y. The current chip position is located in the ath column from left to right and the bth row from bottom to top of a mounting rack on a certain side of the intelligent computing center. The coordinates of the current chip position are X. a ,Y b , the chip position coordinates adjacent to it are X a+1 ,Y b , X a ,Y b+1 , X a―1 ,Y b , X a ,Y b―1 , the adjacent chip positions located at the peripheral chip positions need to subtract the non-existent coordinates;

[0046] S5-2. When the GPU computing chip at the adjacent position of the current chip is working, it will emit heat and cause a temperature rise effect Q0=εt on the current GPU computing chip, where ε is the heat accumulation coefficient caused by the continuous operation of the GPU computing chip at the adjacent position, and t is the working time of the GPU computing chip at the adjacent position. The temperature rise effect caused by all the surrounding GPU computing chips is Where j is the number of GPU computing chips adjacent to a GPU computing chip, so the corrected The adjacent temperature rise is taken into account in the coolant pumping flow rate through calculation, so that the coolant pumping can more accurately match the current working environment.

[0047] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0048] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. Cold plate liquid cooling module for intelligent computing center, characterized by: The invention comprises a water pump (2) and a liquid cooling shell (6), wherein the outlet end of the water pump (2) is connected to a main control valve (1), and the outlet end of the main control valve (1) is connected to a plurality of two-position two-way reversing valves (4), wherein the first port of the two-position two-way reversing valve (4) is connected to the main control valve (1), and the second port of the two-position two-way reversing valve (4) is connected to the liquid cooling shell (6), and one end of the liquid cooling shell (6) is connected to the third port of the two-position two-way reversing valve (4), and the fourth port of the two-position two-way reversing valve (4) is connected to a refrigerator (3), and one end of the refrigerator (3) is connected to the water pump (2), and the liquid cooling shell (6) is mounted on a GPU computing chip.

2. The cold plate liquid cooling module for an intelligent computing center according to claim 1, characterized in that: A flow meter (8) is connected between the fourth port of the two-position two-way reversing valve (4) and the refrigerator (3); a throttle valve (5) and a diverter valve (7) are respectively connected at both ends of the liquid cooling shell (6); one end of the diverter valve (7) is connected to the refrigerator (3); and one end of the throttle valve (5) is connected to the second port of the two-position two-way reversing valve (4).

3. The cold plate liquid cooling module for the intelligent computing center according to claim 2, characterized in that: The control system of the liquid cooling module comprises a flow calculation module, a flow shunt control module, a valve reversing module, a GPU state monitoring module, and a heat generation estimation module, wherein the flow calculation module is electrically connected to the throttle valve (5), the flow shunt control module is electrically connected to the flow shunt valve (7), the valve reversing module is electrically connected to the two-position two-way reversing valve (4), and the GPU state monitoring module is electrically connected to the heat generation estimation module and the flow calculation module; The flow calculation module is used to calculate the flow of coolant passing through each liquid cooling shell (6), the flow distribution control module is used to control the flow distribution valve (7) to distribute the flow of coolant flowing out of the liquid cooling shell (6), the valve reversing module is used to control each two-position two-way reversing valve (4) to perform reversal, the GPU status monitoring module is used to monitor the working status of the GPU computing power chip, and the heat generation estimation module is used to estimate the heat generation according to the working status of the GPU computing power chip, thereby obtaining the coolant flow that needs to be pumped to the liquid cooling shell (6).

4. The cold plate liquid cooling module for an intelligent computing center according to claim 3, characterized in that: The working method of the control system of the liquid cooling module is: S1, start the water pump (2) to mobilize the flow of the coolant, at which time the main control valve (1) is opened, and the coolant enters from the first port of the two-position two-way reversing valve (4), flows from the second port to the liquid cooling shell (6), flows out of the liquid cooling shell (6), enters the third port, and is discharged from the fourth port to the refrigerator (3) to pump the coolant forward; S2, when reversing, the valve reversing module controls each two-position two-way reversing valve (4) to switch the flow direction of the coolant, at which time the coolant enters from the first port of the two-position two-way reversing valve (4), is discharged from the third port and flows to the liquid cooling shell (6), and then enters the second port after being discharged from the liquid cooling shell (6) and enters the two-position two-way reversing valve (4), and is discharged from the fourth port to the refrigerator (3), and the coolant is pumped in the reverse direction; S3, obtaining the real-time working status of the GPU computing chip through the GPU status monitoring module, monitoring the computing volume and GPU utilization rate of the GPU computing chip in artificial intelligence training, calculating the cooling liquid flow required by each liquid cooling shell (6) matching each GPU computing chip, and realizing flow control through the throttle valve (5); S4. When the coolant is pumped in the forward direction, the opening of each diverter valve (7) is adjusted so that the excess coolant flows back to the refrigerator (3) through the diverter valve (7). When the coolant is pumped in the reverse direction, the opening of the throttle valve (5) is maximized, and the coolant is quickly pumped in the reverse direction under the condition of maximum flow, and the forward pumping state is returned after a period of time.

5. The cold plate liquid cooling module for an intelligent computing center according to claim 4, characterized in that: In S3, the specific calculation method of the coolant flow rate is: S3-1. Enable the API tool in the GPU status monitoring module to obtain the working status data of each GPU in real time. First, record the current computing amount being processed, obtain it through the number of floating-point operations k processed per second, and monitor the GPU utilization rate at the same time to obtain the real-time GPU utilization rate n%. Based on these two indicators, the heat generation is calculated using the formula Q=αk+lgβ(n%+1), where α is the weight coefficient of the number of floating-point operations, and β is the weight coefficient of the GPU utilization rate; S3-2. According to the estimated calorific value and the specific heat capacity of the coolant, the required coolant flow rate is calculated. The formula is coolant flow rate L = μQ, where μ is the specific heat capacity coefficient of the coolant. The coolant flow rate L of each liquid-cooled shell (6) is calculated independently to obtain a coolant flow rate set {L1, L2, ..., L x }, where x is the number of liquid cooling shells (6).

6. The cold plate liquid cooling module for an intelligent computing center according to claim 5, characterized in that: In S4, the specific method for adjusting the opening of each diverter valve (7) is: first determine the pumping flow rate L0 of the water pump (2), and then adjust the opening of each diverter valve (7) according to the coolant flow rate set {L1, L2, ..., L x }Calculate the actual total coolant flow ∑L x , then calculate the average flow rate of each liquid cooling shell (6) The flow rate divided by each diverter valve (7) is where i∈[1~x], L i is the coolant flow rate of the liquid cooling shell (6) with serial number i.

7. The cold plate liquid cooling module for an intelligent computing center according to claim 6, characterized in that: The control system of the liquid cooling module also includes a heat impact correction module, which is used to correct the cooling liquid flow of the liquid cooling shell (6) according to the heat impact of the GPU computing chip at the adjacent position. The heat impact correction module includes a coordinate recording module, a working time recording module, an adjacent coordinate judgment module, and a heat accumulation calculation module. The coordinate recording module is used to record the coordinate position of each liquid cooling shell (6), the working time recording module is used to record the working time of each GPU computing chip, the adjacent coordinate judgment module is used to judge the number of GPU computing chips adjacent to each GPU computing chip, and the heat accumulation calculation module is used to calculate the cumulative impact of heat according to the number of adjacent computing chips and the working time.

8. The cold plate liquid cooling module for an intelligent computing center according to claim 7, characterized in that: The working method of the heating effect correction module is as follows: S5-1. When recording the coordinates of the adjacent chip positions of a certain chip position, let the horizontal coordinate parameter of the chip position be X, and the vertical coordinate parameter be Y. The current chip position is located in the ath column from left to right and the bth row from bottom to top of a mounting rack on a certain side of the intelligent computing center. The coordinates of the current chip position are calculated as (X a ,Y b ), and the chip position coordinates adjacent to it are (X a+1 ,Y b )、(X a ,Y b+1 )、(X a―1 ,Y b )、(X a ,Y b―1 ), the adjacent chip positions located at the peripheral chip positions need to subtract the non-existent coordinates; S5-2. When the GPU computing chip at the adjacent position of the current chip is working, it will emit heat and cause a temperature rise effect Q0=εt on the current GPU computing chip, where ε is the heat accumulation coefficient caused by the continuous operation of the GPU computing chip at the adjacent position, and t is the working time of the GPU computing chip at the adjacent position. The temperature rise effect caused by all the surrounding GPU computing chips is Where j is the number of GPU computing chips adjacent to a GPU computing chip, so the corrected

Images