Temperature prediction system and method for a server's PCIe chip

Abstract

The application provides a temperature prediction method for a PCIe chip of a server, comprising defining a temperature prediction model for the PCIe chip using a gating recurrent unit of a recurrent neural network, the temperature prediction model comprising an input end and an output end, collecting training data of the temperature prediction model according to response changes between a plurality of control variables, training the temperature prediction model using the training data at the input end, so as to obtain a training result close to a measured temperature of the PCIe chip from the output end, evaluating the training result to obtain a plurality of features that can best reflect temperature changes of the PCIe chip, performing error analysis on the training result to obtain a group of key features from the plurality of features, forming a temperature predictor according to the key features and the temperature prediction model, and generating a predicted temperature of the PCIe chip through the temperature predictor. The temperature prediction method can solve the problem that the chip on the PCIe card cannot return the temperature value.

Description

Technical Field

[0001] This invention relates to a method for predicting server temperature, and more particularly to a method for predicting server temperature using a neural network model. Background Technology

[0002] A server's heat dissipation capability directly affects its performance and stability. Therefore, servers are equipped with a baseboard management controller (BMC) that manages the server's operating environment. The BMC is responsible for processing information such as power supply, voltage, temperature, and fan speed. After reading the temperature of each component, the BMC controls the server's fan speed according to different conditions to ensure timely heat dissipation.

[0003] Most internal server components, such as the central processing unit (CPU) and memory (DIMM), can report their own temperature to the management chip. However, many high-speed serial computer bus PCIe (PCI express) cards cannot report their own temperature. PCIe card manufacturers only specify the card's inlet temperature as the basis for heat dissipation during server design and manufacturing. Therefore, server thermal engineers typically place an ambient temperature sensor in front of the PCIe card to detect its inlet temperature. The fan speed is then adjusted based on this ambient temperature. However, relying solely on the inlet temperature for cooling the PCIe card's chips presents two major problems when designing fan control. First, taking a PID (Proportional-Integral-Derivative) controller as an example, if a PID controller is used to regulate fan speed based on the PCIe card's inlet temperature, the temperature control setpoint must be much lower than the PCIe card's inlet temperature. Even if PID feedback control can keep the inlet temperature at the set value, if the fan is running at a low speed, insufficient airflow to the PCIe card will cause the chip on the card to overheat. Servers often experience system crashes due to overheating of the PCIe card chip. Therefore, in practical applications, PCIe cards typically employ a conservative high-speed strategy to ensure their heat dissipation requirements. Secondly, PCIe cards are usually located downstream of the server's fan cooling system. When reading the inlet temperature for fan control, because the sensor itself is downstream, there is a time delay in controlling the fan speed based on the sensing results. Therefore, more time is needed to adjust the controller parameters to achieve better control results.

[0004] To prevent overheating of the chips on the card, thermal engineers typically connect thermocouple sensors to the PCIe card chips during the design phase. These thermocouples monitor the chip temperature, and after repeated modifications to control parameters and verification, the safety of the chip temperature is confirmed. Under the most stringent conditions (worse case), the PCIe card's input temperature is usually set at 40°C or lower. As a result, to meet the PCIe card's input temperature requirements, the system fan often runs at high speed, wasting excessive energy. Summary of the Invention

[0005] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a temperature prediction system and method for PCIe chips in servers, which solves the problem that the chips on PCIe cards cannot transmit temperature data in the prior art.

[0006] To achieve the above and other related objectives, a first aspect of the present invention provides a method for predicting the temperature of a PCIe chip in a server, comprising: defining a temperature prediction model for a PCIe chip using a gated recurrent unit (GRU) of a recurrent neural network, the temperature prediction model including an input and an output; collecting training data of the temperature prediction model based on the response changes between a plurality of control variables; training the temperature prediction model at the input using the training data to obtain a training result from the output that approximates a measured temperature of the PCIe chip; evaluating the training result to obtain a plurality of features that best reflect the temperature changes of the PCIe chip; performing an error analysis on the training result to obtain a set of key features from the plurality of features; forming a temperature predictor based on the key features and the temperature prediction model; and generating a predicted temperature of the PCIe chip using the temperature predictor.

[0007] In one embodiment of the first aspect, the control variables include: the power of a PCIe card, including two states: on and off; the utilization rate of a processor, including five states: idle, 25% utilization, 50% utilization, 75% utilization, and 100% utilization; the speed of a server fan, including eight states: 30% speed, 40% speed, 50% speed, 60% speed, 70% speed, 80% speed, 90% speed, and 100% speed; and the intake air temperature of a server, including 18°C ​​to 25°C.

[0008] In one embodiment of the first aspect, the training data includes the processor utilization rate, the server fan speed, the PCIe card power, and the measured temperature of the PCIe chip.

[0009] In one embodiment of the first aspect, the measured temperature is obtained by a thermocouple disposed on the PCIe chip.

[0010] In one embodiment of the first aspect, the plurality of features include any combination of the power of a PCIe card, the speed of a server fan, the intake temperature of a server, the temperature of a processor, and the inlet temperature of a PCIe card; and the key features include the power of the PCIe card, the speed of the server fan, the temperature of the processor, and the inlet temperature of the PCIe card.

[0011] In one embodiment of the first aspect, the error analysis is a root mean square error analysis.

[0012] In one embodiment of the first aspect, the temperature prediction method for the PCIe chip of the server further includes controlling the speed of a server fan based on the predicted temperature of the PCIe chip.

[0013] A second aspect of the present invention provides a temperature prediction system for a PCIe chip in a server, comprising: a server including a PCIe chip and a server fan; a temperature predictor including: a temperature prediction model for the PCIe chip in the server defined by a gated recurrent unit of a recurrent neural network; and a set of key features that best reflect the temperature changes of the PCIe chip; and a baseboard management controller for controlling the temperature prediction model to generate a predicted temperature of the PCIe chip based on the key features, and controlling the speed of the server fan based on the predicted temperature.

[0014] In one embodiment of the second aspect, the key features include the power of the PCIe card, the speed of the server fan, the temperature of the processor, and the inlet temperature of the PCIe card.

[0015] In summary, conservative temperature control of server PCIe card chips often involves running fans at high speeds to prevent overheating, primarily because the chips on the PCIe card cannot report their temperature readings. Therefore, this invention proposes a method and system design for predicting the temperature of chips on a PCIe card. This allows for direct design of a controller targeting the chip's temperature, addressing the issue of the chip's inability to report temperature values. Furthermore, the temperature prediction system not only obtains the current chip temperature but also predicts future temperature trends, thus resolving the time delay issue in fan speed response. Attached Figure Description

[0016] Figure 1 The diagram shown is a schematic of the temperature prediction system of the PCIe chip of the server in an embodiment of the present invention.

[0017] Figure 2 The diagram shown is a schematic representation of the temperature prediction model in an embodiment of the present invention.

[0018] Figure 3 Displayed as Figure 1 Another schematic diagram of the temperature prediction system of the PCIe chip in the server shown.

[0019] Figure 4 Displayed as Figure 2 The diagram shows the training process of the temperature prediction model.

[0020] Figure 5 Displayed as Figure 2 The diagram shows a prediction from the temperature prediction model.

[0021] Component designation explanation

[0022] 100 Server Temperature Prediction Device

[0023] 200 Temperature Prediction Model

[0024] 2. Central Processing Unit

[0025] 4. Memory

[0026] 6 Hard disk modules

[0027] 8 Fan Module

[0028] 10 Power Supply

[0029] 12 PCIe cards

[0030] 20 Baseboard Management Controller

[0031] 30 servers

[0032] T amb Intake temperature

[0033] T CPU Central Processing Unit Temperature

[0034] T in PCIe card inlet temperature

[0035] T PCIE PCIe card chip temperature

[0036] P PCIe card chip power

[0037] U fan speed Detailed Implementation

[0038] Figure 1This is a schematic diagram of a server PCIe chip temperature prediction system 100 in an embodiment of the present invention. The server PCIe chip temperature prediction system 100 includes a server 30 and a baseboard management controller 20. The server 30 includes a central processing unit 2, memory 4, a hard disk module 6, a fan module 8, a power supply 10, and a PCIe card 12. The baseboard management controller 20 controls the temperature prediction model to generate a predicted temperature for the PCIe card 12 chip based on key characteristics, and controls the server fan speed based on the predicted temperature.

[0039] The server's PCIe chip temperature prediction system 100 also includes a temperature predictor. The temperature predictor includes a temperature prediction model for the server's PCIe card 12 chip defined by gated recurrent units of a recurrent neural network, and a set of key features that best reflect temperature changes in the PCIe card 12 chip. The temperature prediction model and the set of key features can be stored in memory and executed by a processor. The memory and processor can be of any form.

[0040] Please refer to Figure 2 and Figure 3 . Figure 2 This is a schematic diagram of the temperature prediction model 200 in an embodiment of the present invention. Figure 3 This is another schematic diagram of the temperature prediction system 100 of the PCIe chip of the server in this embodiment of the invention. This embodiment uses a gated recurrent unit (GRU) in a recurrent neural network (RNN) as the architecture of the temperature prediction model 200. Because recurrent neural networks have the ability to memorize historical data, this deep learning architecture can effectively predict future trends from historical data. The goal of the temperature prediction model 200 is to infer output data y(k), y(k+1), y(k+2)... from the known input training data x1, x2... . k is the sampling point, and the sampling period is 1 second, but not limited to this. The selection of training data has an absolute impact on the accuracy of the prediction system. The embodiment selects the following training data: the intake air temperature T of the server 30. amb The fan speed of fan module 8 and the temperature T of central processing unit 2. CPU The power P of PCIe card 12 and the inlet temperature T of PCIe card 12 in Additionally, the output data of the temperature prediction model is the chip temperature T of the PCIe card 12. PCIE The generation, storage, and processing of training data, as well as the training and evaluation of models, can be automated through programs.

[0041] Table 1

[0042]

[0043]

[0044] Referring to the control variables in Table 1, the control range modulation is for illustrative purposes only and is not intended to limit the invention. The control variables can be used to generate input data for the predictive model. The chip power P of PCIe card 12 has two states: ON and OFF. The control signal for fan speed U is a pulse-width modulation (PWM) signal, including eight states: 30%, 40%, 50%, 60%, 70%, 80%, 90%, and 100%. The utilization rate of CPU 2 includes five states: Idle, 25%, 50%, 75%, and 100%, which mainly affects the heat source of the downstream PCIe card 12. In this embodiment, the fan speed, the chip power P of PCIe card 12, and the utilization rate of CPU 2 can be controlled programmatically, and the server intake air temperature T can be detected. amb The temperature of the central processing unit 2 (T) CPU and the chip temperature T of PCIe card 12 PCIE The temperature prediction model 200 is trained by analyzing the changes in temperature. During the design phase of server 30, a thermocouple sensor can be used to sense the chip of PCIe card 12, thereby obtaining the chip's temperature. After training, although the actual chip on PCIe card 12 does not have a thermocouple sensor, the chip temperature T can still be predicted using the temperature prediction model 200 in this embodiment. PCIE The changes.

[0045] Table 2

[0046]

[0047] Table 2 presents the error analysis of the training results under various input features. The error data are illustrative of the experimental results based on this invention and are not intended to limit the invention. 'o' represents the use of the feature, and 'x' represents the non-use of the feature. The chip power P and fan speed U of the PCIe card are key features. From the root mean square error (RMSE) analysis, adding T... CPU T in , and T CPU This can produce a relatively small error range (Group 4 input features). Therefore, the embodiment selects PCIe card chip power P, fan speed U, and CPU temperature T. CPU PCIe card inlet temperature T inThis is a key feature of the temperature predictor. However, the invention is not limited thereto; in another embodiment, the key feature may be any one or a combination of the features in Table 2.

[0048] Figure 4 This is a schematic diagram illustrating the training of the temperature prediction model 200 in an embodiment of the present invention. In this embodiment, the central processing unit is in an idle state, and the control variables in Table 1 are used to train the temperature prediction model 200. When the chip of PCIe card 12 is in an ON state, the chip power of PCIe card 12 is 100%, and the chip temperature T PCIE The power consumption of PCIe card 12 increases when the chip is in the OFF state. The chip power consumption is 0%, and the chip temperature T... PCIE Then it decreases. The temperature T of the central processing unit. CPU The fan speed U during training varies depending on the on / off state of the PCIe card 12 chip. The training data generated by the control variables and other parameters in this embodiment can be used to train the temperature prediction model 200 so that the model's output data, i.e., the chip temperature TPCIE, can closely approximate the measured temperature.

[0049] Figure 5 This is a schematic diagram of the temperature predictor in an embodiment of the present invention. The temperature predictor is formed by the key feature temperature prediction model 200. As can be seen from the figure, when the fan speed of the fan module 8 gradually increases from 40% to 80%, the central processing unit temperature TCPU and the server intake air temperature T... amb All changes were minor, however, the chip temperature T of the PCIe card 12 increased. PCIE When the PCIe 12 chip is ON, its temperature decreases as the fan speed increases. Furthermore, the PCIe 12 chip temperature T... PCIE The actual value is very close to the predicted value, proving that the temperature predictor can accurately predict the chip temperature T of the PCIe card 12. PCIE .

[0050] In summary, the embodiments provide a temperature prediction system and method for a server's PCIe chip. This includes defining training data and output data for a temperature prediction model of the server's PCIe chip; using the training data to train and test the temperature prediction model; adjusting the temperature prediction model so that its output data approximates the measured value; and using a temperature predictor formed by the temperature prediction model and key features to predict the temperature of the server's PCIe card chip. This allows for the prediction of temperature trend changes in the server's PCIe card chip and solves the time delay problem in fan speed response.

[0051] In one embodiment of the present invention, the temperature prediction system and method for the PCIe chip of the server described in the present invention can be applied to a server that can be used for artificial intelligence (AI) computing, edge computing, or as a 5G server, cloud server, or vehicle networking server.

[0052] The above description is only a preferred embodiment of the present invention. All equivalent changes and modifications made within the scope of the claims of the present invention should be included in the scope of the present invention.

Claims

1. A method for predicting the temperature of a server's PCIe chip, characterized in that, Include: A temperature prediction model for a PCIe chip is defined using a gated recurrent unit of a recurrent neural network, the temperature prediction model including an input and an output. Training data for the temperature prediction model is collected based on the changes in the responses of a plurality of control variables to each other. The temperature prediction model is trained using the training data at the input terminal to obtain a training result from the output terminal that approximates a measured temperature of the PCIe chip. The training result is evaluated to obtain a plurality of features that best reflect the temperature change of the PCIe chip. An error analysis is performed on the training results to obtain a set of key features from the plurality of features; A temperature predictor is formed based on the key features and the temperature prediction model; as well as A predicted temperature of the PCIe chip is generated by the temperature predictor.

2. The temperature prediction method for the PCIe chip of a server according to claim 1, characterized in that, The control variables include: The power of a PCIe card includes two states: on and off. The utilization rate of a processor includes five states: idle, 25% utilization, 50% utilization, 75% utilization, and 100% utilization. A server fan's speed includes eight states: 30%, 40%, 50%, 60%, 70%, 80%, 90%, and 100%. as well as The intake air temperature of a server ranges from 18°C ​​to 25°C.

3. The temperature prediction method for the PCIe chip of a server according to claim 2, characterized in that: The training data includes the processor utilization rate, the server fan speed, the PCIe card power, and the measured temperature of the PCIe chip.

4. The temperature prediction method for the PCIe chip of a server according to claim 3, characterized in that: The measured temperature is obtained by a thermocouple disposed on the PCIe chip.

5. The temperature prediction method for the PCIe chip of a server according to claim 3, characterized in that: The plurality of features includes any combination of the following: the power of a PCIe card, the speed of a server fan, the intake air temperature of a server, the temperature of a processor, and the inlet temperature of a PCIe card; and The key features include the power of the PCIe card, the speed of the server fan, the temperature of the processor, and the inlet temperature of the PCIe card.

6. The temperature prediction method for the PCIe chip of a server according to claim 1, characterized in that: The error analysis is a root mean square error analysis.

7. The temperature prediction method for the PCIe chip of a server according to claim 1, characterized in that: The method for predicting the temperature of the PCIe chip in the server also includes controlling the speed of a server fan based on the predicted temperature of the PCIe chip.

8. A temperature prediction system for a server's PCIe chip, characterized in that, Include: A server, comprising a PCIe chip and a server fan; A temperature predictor, comprising: A temperature prediction model for the PCIe chip of the server, defined by a gated recurrent unit of a recurrent neural network; and The set of key features that best reflects the temperature changes of the PCIe chip; as well as A baseboard management controller is used to control the temperature prediction model to generate a predicted temperature of the PCIe chip based on the key characteristics, and to control the speed of the server fan based on the predicted temperature.

9. The temperature prediction system for the PCIe chip of a server according to claim 8, characterized in that: The key features include the power of the PCIe card, the speed of the server fan, the temperature of the processor, and the inlet temperature of the PCIe card.

Images