Power allocated between the accelerated processing unit and the discrete graphics processing unit.
By dynamically adjusting power allocation through the SMU, the heat dissipation problem of heterogeneous processing units and discrete graphics processing units is solved, achieving efficient heat dissipation management and performance optimization, thereby improving user experience and device lifespan.
Patent Information
- Application Number
- CN202080085571.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2019-12-12
- Filing Date
- 2020-12-11
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2040-12-11
AI Technical Summary
In high-performance computing devices, it is difficult to balance the heat dissipation requirements of heterogeneous processing units and discrete graphics processing units, leading to increased temperatures, which affects device lifespan and user experience, especially in ultra-thin platforms where performance improvements are limited.
Power allocation is dynamically adjusted through the System Management Unit (SMU) to achieve power sharing between integrated coprocessors and discrete coprocessors based on workload characteristics and thermal constraints, and power is dynamically transferred to optimize heat dissipation.
It effectively controls the surface temperature of computing devices, improves performance, extends device life, provides users with a comfortable operating experience, and optimizes performance under different workloads.
Smart Images

Figure CN114787751B_ABST
Abstract
Description
Background Technology
[0001] High-performance computing devices such as laptops, tablets, mobile phones, and smartwatches include heterogeneous processing units such as accelerated processing units (APUs), which include various types of coprocessors. For example, an APU typically includes one or more central processing unit (CPU) cores and one or more graphics processing unit (GPU) cores, sometimes referred to as integrated GPUs (iGPUs). Additional graphics processing capabilities are provided by including discrete graphics processing units (dGPUs) in the computing device, and in some cases, general-purpose computing capabilities. The power dissipated in APUs and dGPUs tends to raise the temperature of the computing device. Therefore, APUs and dGPUs are connected to cooling systems such as heat pipes, which transfer heat away from the APUs and dGPUs towards corresponding heatsinks that dissipate the heat energy into the environment. The cooling requirements of the system are determined at least in part by the architecture of the computing device and the heat dissipated by the APUs and dGPUs. The cooling systems for APUs and dGPUs keep their temperatures below levels that could damage components or reduce their lifespan. Attached Figure Description
[0002] This disclosure will be better understood by referring to the accompanying drawings, which will make its many features and advantages readily apparent to those skilled in the art. The same reference numerals are used in different drawings to indicate similar or identical items.
[0003] Figure 1 It is a block diagram of an apparatus including an integrated processing unit and a separate discrete coprocessor according to some implementation schemes.
[0004] Figure 2 It is a block diagram of a processing system, according to some implementation schemes, including heat dissipation mechanisms for integrated and discrete coprocessors.
[0005] Figure 3 The diagram illustrates, based on some implementation schemes, the dynamic transfer of shared platform power from integrated coprocessors to discrete coprocessors.
[0006] Figure 4 The diagram illustrates, based on some implementation schemes, the dynamic transfer of shared platform power from discrete coprocessors to integrated coprocessors.
[0007] Figure 5 This is a flowchart illustrating a method for modifying power allocation between integrated and discrete coprocessors based on workload characteristics and platform power limitations in some implementation schemes. Detailed Implementation
[0008] The System Management Unit (SMU) implements power management techniques to allocate power to integrated coprocessor cores (such as the Central Processing Unit (CPU) cores in an Accelerated Processing Unit (APU)) and discrete coprocessor cores (such as discrete graphics processing units (dGPUs)) based on corresponding fixed processor power and thermal envelopes. For example, the SMU allocates power to the CPU cores in an APU based on Thermal Design Power (TDP), which is set based on the worst-case scenario of running heavy workloads on the APU. TDP represents the upper limit of sustainable power and is used to determine system cooling requirements. In another example, the SMU allocates power to the dGPU based on Total Graphics Power (TGP), which represents the maximum amount of graphics board power supplied by the system power supply to the dGPU. In some cases, the device's maximum operating temperature is more limited by user perception than by silicon temperature. Heat generated by the processing units in a handheld device is conducted to the device's outer surfaces, such as the display and casing, where the user interacts with the device during operation. To provide a comfortable experience for users, the maximum power budget allocated to the APU and dGPU is set below the corresponding fixed limits of TDP or TGP, respectively, to keep the surface temperature of the device below a value that would make users feel uncomfortable from the heat.
[0009] For ultra-thin platforms used in handheld or wearable computing devices, the chassis's thermal capacity and cooling solutions cannot support providing the APU at full TDP while simultaneously providing the dGPU and associated video memory at full TGP. Running the APU at full TDP and the dGPU at full TGP would heat the ultra-thin platform to temperatures that could damage the device or at least cause user discomfort. Therefore, the power supplied to the APU is artificially and statically limited to a lower level that allows the dGPU to operate at TGP while the APU operates at its (reduced) maximum power level. Furthermore, the power supplied to the dGPU does not increase in response to the APU operating below its maximum power level, nor does it increase in response to the dGPU operating below TGP. For high-performance platforms, the chassis's thermal capacity and cooling solutions can support providing the APU at full TDP while simultaneously providing the dGPU and associated video memory at full TGP. However, the opportunity for performance gains is limited under core-intensive workloads that primarily utilize the APU or graphics-intensive workloads that primarily utilize the dGPU.
[0010] Figures 1 to 5Implementations of computing devices are disclosed that allocate power to one or more integrated coprocessor cores and one or more discrete coprocessor cores based on the characteristics of workloads executed on integrated and discrete coprocessor cores and platform power constraints shared by the integrated and discrete coprocessor cores. In some implementations, the power allocation circuitry manages the power consumption of the APU implementing the integrated coprocessor cores and the dGPU implementing the discrete coprocessor cores based on available platform power (e.g., supplied by AC power or battery) and thermal constraints. For example, the power allocation circuitry selectively allocates power away from the APU and towards the dGPU for graphics-intensive workloads. The amount of power supplied to the dGPU is determined based on available platform power and thermal constraints such as the maximum surface temperature of the computing device. The surface temperature is determined based on pre-calibrated parameters that determine the relationship between temperatures measured using one or more temperature sensors and the surface temperature of the computing device. The power allocation circuitry also dynamically modifies the power supplied to the integrated and discrete coprocessor cores in response to changes in the characteristics of workloads executed on the integrated and discrete coprocessor cores and changes in surface temperature. For example, the power allocation circuitry shifts power from the dGPU to the APU in response to a workload changing from graphics-intensive to core-intensive. In another example, the power distribution circuit reduces the power supplied to the dGPU for graphics-intensive workloads in response to a measured temperature rise exceeding a threshold. Some implementations of the power distribution circuit allocate power to the APU based on the maximum power dissipation (Qmax) of the heat pipes associated with the APU.
[0011] Figure 1 This is a block diagram of an apparatus 100 including an integrated processing unit 105 and a separate discrete coprocessor 110 according to some embodiments. In the illustrated embodiment, the integrated processing unit 105 is implemented as an accelerated processing unit (APU) 105, and the separate discrete coprocessor 110 is implemented as a graphics processing unit (dGPU) 110. However, in other embodiments, the integrated processing unit 105 or the discrete coprocessor 110 is implemented using other types of coprocessors, digital signal processors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), etc. Some embodiments of the apparatus 100 are embodied in handheld or wearable devices, such as laptop computers, handheld computers, tablet computers, mobile devices, telephones, personal data assistants (“PDAs”), music players, gaming devices, etc.
[0012] The APU 105 includes integrated coprocessor cores such as one or more central processing unit (CPU) cores 115 and one or more graphics processing unit (GPU) cores 120, collectively referred to herein as integrated GPU (iGPU) 125. One or more CPU cores 115 and one or more iGPUs 125 are disposed on the same integrated circuit (IC) die or on different IC dies within the same IC package. The CPU core 115 has an independently controlled power layer that allows voltage and frequency control independent of the voltage and frequency associated with the GPU core 120. Some embodiments of the dGPU 110 include one or more discrete coprocessor cores 127. The dGPU 110 also has an independently controlled power layer that allows voltage and frequency control of the voltage and frequency supplied to the dGPU 110 (or discrete coprocessor core 127) independent of the voltage and frequency associated with the APU 105, CPU core 115, or GPU core 120.
[0013] Device 100 includes system memory 130, display 135, and power supply 140 (including voltage regulator, battery, and battery charging device—not shown separately for clarity). Although not shown separately for clarity... Figure 1 As shown, the northbridge controller in APU 105 provides an interface to system memory 130. The operation of device 100 is typically controlled by an operating system that includes software that interacts with the various components of device 100. APU 105 integrates CPU core 115 and GPU core 120 on a common semiconductor die, allowing them to share on-die resources such as memory hierarchy and interconnects.
[0014] Device 100 includes a housing 145 that supports a display 135 and surrounds active components of device 100. Housing 145 also provides an outer surface along which a user interacts with device 100. In implementations where display 135 is a touchscreen, APU 105 controls and receives user input from display 135. dGPU 110 or iGPU 125 provides signals to display 135 via bus 150, such as a peripheral component interconnect (PCI, PCI-E) bus. These signals include information representing the colors and intensities produced by pixels in display 135, which are combined to produce an image seen by a user viewing display 135. If display 135 is a touchscreen, bus 150 is also used to transmit signals from display 135 to APU 105, which initiates operation based on the location of a touch point on display 135 (and, in some cases, based on the amount or duration of pressure applied by the user).
[0015] One or more temperature sensors 151, 152 are provided in device 100, such as sensor 151 near housing 145 and sensor 152 near display 135. A surface temperature is estimated based on the temperature measured by temperature sensors 151, 152, indicating the temperature perceived by a user holding the device. In some embodiments, the relationship between the measured temperature and the surface temperature is pre-calibrated by comparing the actual measured surface temperature with the measured values. For example, model parameters relating the surface temperature to the measured temperature are determined using a calibration process that directly measures the temperature of portions of housing 145 and correlates these measurements with the temperatures measured by sensors 151, 152. In some embodiments of device 100, activity counters such as CPU counter 155, GPU counter 160, and other counters such as memory counters (not shown for clarity) are provided. Figure 1 (As shown in the figure) to measure the activity of the device that generates the components, in order to estimate the heat they generate and its effect on the surface temperature.
[0016] Power is distributed from power supply 140 to APU 105 and dGPU 110 via power distribution circuitry 165. In the illustrated embodiment, APU 105 includes SMU 170, and dGPU 110 includes SMU 175. SMU 175 monitors power consumption characteristics in dGPU 110, such as the current power supplied to or dissipated in dGPU 110. SMU 175 provides APU 105 with information indicating power consumption characteristics. Power consumption information is provided periodically at predetermined time intervals in response to power-related events in dGPU 110 or at other times. Based on the received power consumption information, SMU 170 determines and dynamically adjusts the power supplied from power supply 140 to APU 105 and dGPU 110 based on the characteristics of the workloads performed on APU 105 and dGPU 110. Power allocation to the APU 105 and dGPU 110 is also based on a platform power limit shared by both the APU 105 and dGPU 110. In some implementations, the platform power limit is equal to (or determined based on) the sum of the thermal design power (TDP) and the total graphics power (TGP), where TDP is set based on a worst-case scenario of running a heavy workload on the APU 105, and TGP represents the maximum amount of graphics board power supplied to the dGPU 110. TDP represents the upper limit of sustainable power supplied to the APU 105 and is used to determine system cooling requirements. In some implementations, power supplied to the dGPU 110 exceeds the TGP without causing any damage to the dGPU 110 or other circuitry within device 100.
[0017] The SMU 170 modifies the power allocated to the APU 105 and dGPU 110 in response to changes in the characteristics of the workload performed on the APU 105 or dGPU 110. Some implementations of the SMU 170 shift power from the APU 105 to the dGPU 110 in response to a change in workload characteristics from a core-intensive workload primarily consuming APU 105 resources to a graphics-intensive workload primarily consuming dGPU 110 resources. The SMU 170 also shifts power from the dGPU 110 to the APU 105 in response to a change in workload characteristics from a graphics-intensive workload primarily consuming dGPU 110 resources to a core-intensive workload primarily consuming APU 105 resources.
[0018] Some implementations of the SMU 170 implement surface temperature tracking (STT). For example, the SMU 170 may implement an STT controller 172 that sets dynamic power limits for the device 100 based on a determined surface temperature. The SMU 170 uses information provided by the STT controller 172, combined with constraints imposed by platform power limits and the characteristics of the workloads executed on the APU 105 and dGPU 110, to dynamically determine the power allocation between the APU 105 and dGPU 110. Within the power allocated to the APU 105, the STT controller in the SMU 170 implements dynamic voltage-frequency regulation (DVFS) to adapt to the voltage and clock levels of the CPU core 115 and GPU core 120. Some implementations of the SMU 170 also control the bandwidth allocated to the system memory 130 or the battery charging rate employed by the power supply 140 to control their respective thermal contributions.
[0019] The surface temperature of device 100 is estimated using a pre-calibrated correlation between the temperatures measured by temperature sensors 151 and 152 and the measured surface temperature. The pre-calibrated correlation sets the value of a parameter that correlates the temperatures measured by temperature sensors 151 and 152 to a user-perceived surface temperature using device 100. SMU 170 modifies the power allocated to APU 105 or dGPU 110 based on the estimated surface temperature. For example, if SMU 170 determines that the sum of a first power supplied to APU 105 and a second power supplied to dGPU 110 is less than a platform power limit, SMU 170 increases the power supplied to dGPU 110. Furthermore, SMU 170 (or STT controller 172) determines or modifies the amount of power supplied to dGPU 110 based on a comparison of the surface temperature with a maximum surface temperature set by the thermal constraints of device 100 or dGPU 110. For example, SMU 170 reduces the power supplied to dGPU 110 in response to a surface temperature exceeding the maximum surface temperature. In another example, the SMU 170 increases the power supplied to the dGPU 110 (and in some cases the APU 105) in response to a surface temperature below the maximum surface temperature.
[0020] Figure 2 This is a block diagram of a processing system 200, according to some embodiments, including a heat dissipation mechanism for an integrated coprocessor 205 and a discrete coprocessor 210. In the illustrated embodiment, the integrated coprocessor 205 is implemented as an APU 205, and the discrete coprocessor 210 is implemented as a dGPU 210. The processing system 200 is used to implement... Figure 1 Some embodiments of the illustrated device 100 are shown. The processing system 200 includes a system management unit (SMU) 215 that distributes power from a power supply 220 to the APU 205 and the dGPU 210. Some embodiments of the power supply 220 include (or are connected to) power distribution circuitry to communicate between the APU 205 and the dGPU 210 to exchange workload activity requirements and power limit information with the SMU 215. For example, the GPU 205, dGPU 210, and SMU 215 can communicate via a PCIe bus using a serial bus two-wire (I2C) protocol, a single-wire protocol, or the like.
[0021] In the illustrated implementation, APU 205 is connected to heat pipe 225, which conducts heat from APU 205 to a corresponding heatsink 230. dGPU 210 is connected to heat pipe 235, which conducts heat from dGPU 210 to a corresponding heatsink 240, which may be the same as or different from heatsink 230, depending on the implementation. Some implementations of SMU 215 allocate power to APU 205 based on the maximum power dissipation (Qmax) of heat pipe 225, and allocate power to dGPU 210 based on the Qmax of heat pipe 235.
[0022] Figure 3 Figure 300 illustrates, according to some implementation schemes, the dynamic transfer of shared platform power from an integrated coprocessor to a discrete coprocessor. Figure 300 shows dynamic power allocation performed by power distribution circuitry, such as... Figure 1 The SMU 170 shown is Figure 2 Some implementations of the SMU 215 are shown. The power distribution circuitry allocates power based on the characteristics of the workload executed on the APU and dGPU, as well as the platform power limit 305, which indicates the total power available for the power distribution circuitry to allocate.
[0023] Before time T1, the power distribution circuitry provides a first power 310 to the APU and a second power 315 to the dGPU. In the illustrated embodiment, the device, including the APU and dGPU, is executing workloads that are approximately evenly distributed between core workloads executed on the APU and graphics workloads executed on the dGPU. Therefore, the first power 310 and the second power 315 are approximately equal during the time interval before time T1, although in the illustrated embodiment, the first power 310 is slightly lower than the second power 315. The sum of the first power 310 and the second power 315 is determined based on the platform power limit 305.
[0024] The power distribution circuitry also considers thermal constraints, such as maximum surface temperature, to determine the power supplied to the APU and dGPU. In the illustrated embodiment, the thermal constraints result in less than the total power available under platform power limit 305 being allocated to the APU and dGPU. For example, the total power supplied to the APU and dGPU is equal to the reduced platform power limit 320, which is lower than the platform power limit 305 by an amount 325 determined by the thermal constraints.
[0025] The power distribution circuitry begins transferring power from the APU to the dGPU at time T1. In the illustrated embodiment, the power distribution circuitry transfers power in response to a workload shifting from core-intensive to graphics-intensive. For example, the power distribution circuitry reduces a first power 310 by a first amount and increases a second power 315 by a corresponding amount. Although in the illustrated embodiment the reduction of the first power 310 is equal to the increase of the second power 315, the ratio of the reduction of the first power 310 to the increase of the second power 315 is not always 1:1, and this ratio has different values depending on the power consumption characteristics of the APU and the dGPU. Transferring power from the APU to the dGPU improves the performance of graphics-intensive workloads while still keeping the total distributed power below the platform power limit 305 and below the reduced platform power limit 320.
[0026] Figure 4 Figure 400 illustrates, according to some implementation schemes, the dynamic transfer of shared platform power from a discrete coprocessor to an integrated coprocessor. Figure 400 shows dynamic power allocation performed by power allocation circuitry, such as... Figure 1 The SMU 170 shown is Figure 2 Some embodiments of the SMU 215 are shown. The power distribution circuitry allocates power based on the characteristics of the workloads executed on the APU and dGPU, as well as a platform power limit 405, which indicates the total power available for allocation by the power distribution circuitry. Although the power distribution circuitry considers thermal constraints such as maximum surface temperature to determine the power supplied to the APU and dGPU, in the illustrated embodiment, the thermal constraints do not provide a large constraint on the available power, and the reduced platform power limit 420 is approximately equal to the platform power limit 405.
[0027] Before time T1, the power distribution circuitry provides a first power 410 to the APU and a second power 415 to the dGPU. In the illustrated embodiment, the device, including the APU and dGPU, is performing a workload that is approximately evenly distributed between the core workload performed on the APU and the graphics workload performed on the dGPU. Therefore, the first power 410 and the second power 415 are approximately equal during the time interval before time T1, although in the illustrated embodiment, the first power 410 is slightly lower than the second power 415. The sum of the first power 410 and the second power 415 is determined based on the platform power limit 405.
[0028] The power distribution circuitry begins transferring power from the dGPU to the APU at time T1. In the illustrated embodiment, the power distribution circuitry transfers power in response to a workload shifting from graphics-intensive to core-intensive. For example, the power distribution circuitry increases a first power 410 by a first amount and decreases a second power 415 by a corresponding amount. Although in the illustrated embodiment the increase in first power 410 equals the decrease in second power 415, the ratio of the increase in first power 410 to the decrease in second power 415 is not always 1:1, and this ratio has different values depending on the power consumption characteristics of the APU and the dGPU. In some embodiments, the increase in first power 410 (or the total value of first power 410) is limited to a value below the TDP of the APU. Transferring power from the dGPU to the APU thus improves the performance of core-intensive workloads while still keeping the total distributed power below the platform power limit 405 and below the reduced platform power limit 420.
[0029] Figure 5 This is a flowchart of method 500, which modifies the power allocation for integrated and discrete coprocessors based on workload characteristics and platform power limitations in some implementation schemes. Method 500 in... Figure 1 The SMU 170 in the device 100 shown and Figure 2 Some implementations of the SMU 215 in the processing system 200 shown are implemented. As discussed herein, platform power limits are shared between the APU and dGPU, and the SMU is free to allocate the shared power between the APU and dGPU.
[0030] At box 505, the SMU determines the characteristics of the workload. In some implementations, these characteristics include an indication of the workload's relative core intensity and graphics intensity. Workloads that perform a lot of computation but do not generate images for display are considered more core-intensive, while workloads that perform less computation but generate high-resolution images for display are considered more graphics-intensive.
[0031] At box 510, the SMU collects temperature measurements from one or more sensors distributed throughout the device. In some embodiments, the temperature values measured by the sensors are used to infer thermal constraints (such as surface temperatures) for comparison with corresponding thermal constraints (such as maximum surface temperatures). As discussed herein, thermal conditions are inferred using parameter-defined relationships determined using a calibration process performed during device configuration.
[0032] At decision box 515, the SMU determines whether the workload of the APU or dGPU has changed. Examples of workload changes include, but are not limited to, a shift from core-intensive workload to graphics-intensive workload, a shift from graphics-intensive workload to core-intensive workload, an increase or decrease in the workload allocated to the APU or dGPU, etc. If no change in workload is detected, method 500 flows back to box 505. If a change in workload is detected, method 500 flows to box 520.
[0033] At box 520, the SMU modifies the power allocated to the APU and dGPU based on the modified workload characteristics and platform power limits. In response to a workload shifting from core-intensive to graphics-intensive, allocated power is transferred from the APU to the dGPU. In response to a workload shifting from graphics-intensive to core-intensive, allocated power is transferred from the dGPU to the APU. In some implementations, the modification of power allocation is determined at least in part by thermal constraints such as the maximum surface temperature. For example, the surface temperature of the device is estimated based on temperature sensor measurements and then compared to the highest surface temperature. Power allocation is determined to keep the surface temperature below the maximum surface temperature, for example, by increasing or decreasing the power allocation based on a comparison of the surface temperature and the maximum surface temperature.
[0034] In some embodiments, an apparatus is provided. The apparatus includes an integrated coprocessor comprising at least one central processing unit (CPU) core and at least one graphics processing unit (GPU) core. The integrated coprocessor is configured to generate commands for execution on a discrete coprocessor external to the integrated coprocessor. The apparatus also includes power distribution circuitry configured to selectively provide power to the integrated coprocessor and the discrete coprocessor based on the characteristics of the workloads executed on the integrated coprocessor and the discrete coprocessor, and based on platform power limits shared by the integrated coprocessor and the discrete coprocessor.
[0035] In some implementations, a method is provided. The method includes determining a first power to be provided to an integrated coprocessor, the integrated coprocessor including at least one central processing unit (CPU) core and at least one graphics processing unit (GPU) core. The integrated coprocessor is configured to generate commands for execution on a discrete coprocessor external to the integrated coprocessor. The method also includes determining a second power to be provided to the discrete coprocessor. The method further includes modifying at least one of the first power and the second power based on the characteristics of the workload executed on the integrated coprocessor and the discrete coprocessor and based on platform power limits shared by the integrated coprocessor and the discrete coprocessor.
[0036] In some embodiments, an apparatus is provided. The apparatus includes an integrated coprocessor comprising a central processing unit (CPU) and an integrated graphics processing unit (iGPU). The CPU is configured to generate commands for execution on the iGPU and discrete coprocessors. The apparatus also includes power distribution circuitry configured to dynamically transfer power between the integrated and discrete coprocessors based on platform power limits shared by the integrated and discrete coprocessors and in response to changes in the characteristics of the workloads executed on the integrated and discrete coprocessors.
[0037] Computer-readable storage media include any non-transitory storage medium or combination of non-transitory storage media that can be accessed by a computer system during use to provide instructions and / or data to the computer system. Such storage media include, but are not limited to, optical media (e.g., compressed optical discs (CDs), digital versatile optical discs (DVDs), Blu-ray discs), magnetic media (e.g., floppy disks, magnetic tapes, or magnetic hard disk drives), volatile memory (e.g., random access memory (RAM) or cache), non-volatile memory (e.g., read-only memory (ROM) or flash memory), or microelectromechanical systems (MEMS) based storage media. Computer-readable storage media are embedded in a computing system (e.g., system RAM or ROM), fixedly attached to a computing system (e.g., magnetic hard disk drives), removably attached to a computing system (e.g., optical discs or USB-based flash memory), or coupled to a computer system via a wired or wireless network (e.g., network accessible storage devices (NAS)).
[0038] In some implementations, certain aspects of the techniques described above are implemented by one or more processors of a processing system executing the software. The software includes one or more executable instruction sets stored or otherwise tangibly embodied on a non-transitory computer-readable storage medium. The software includes instructions and certain data that, when executed by one or more processors, manipulate one or more processors to perform one or more aspects of the techniques described above. Non-transitory computer-readable storage media include, for example, disk or optical disk storage devices, solid-state storage devices (such as flash memory, cache, random access memory (RAM), or one or more other non-volatile memory devices), etc. The executable instructions stored on the non-transitory computer-readable storage medium are in the form of source code, assembly language code, object code, or other instruction formats that are interpreted or otherwise executed by one or more processors.
[0039] It should be noted that not all activities or elements described in the general description above are required. A particular activity or part of the apparatus may be unnecessary, and one or more other activities may be performed, or elements other than those described may be included. Furthermore, the order in which the activities are listed is not necessarily the order in which they are performed. Additionally, the corresponding concepts have been described with reference to specific embodiments. However, those skilled in the art will understand that various modifications and changes may be made without departing from the scope of this disclosure as set forth in the appended claims. Therefore, this specification and drawings are to be regarded as illustrative rather than restrictive, and all such modifications are intended to be included within the scope of this disclosure.
[0040] The benefits, other advantages, and solutions to the problems have been described above with reference to specific embodiments. However, the benefits, advantages, and solutions to the problems described, as well as any features that bring about or make more prominent any benefit, advantage, or solution to the problem, should not be construed as being key, essential, or necessary features of any or all claims. Furthermore, the specific embodiments disclosed above are merely illustrative, as the disclosed subject matter can be modified and practiced in different but equivalent ways that will be readily apparent to those skilled in the art who have benefited from the teachings herein. No limitation is intended to be made on the details of the constructions or designs shown herein, except as described in the appended claims. Therefore, it is apparent that the specific embodiments disclosed above can be altered or modified, and all such changes are considered to be within the scope of the disclosed subject matter. Therefore, the protection sought herein is set forth in the following claims.
Claims
1. An apparatus comprising: An integrated coprocessor, comprising a first integrated coprocessor core and a second integrated coprocessor core, the integrated coprocessor being configured to generate commands for execution on a discrete coprocessor external to the integrated coprocessor; and A power distribution circuit is configured to selectively provide power to the integrated coprocessor and the discrete coprocessor based on the characteristics of the workload executed on the integrated coprocessor and the discrete coprocessor and based on a reduced platform power limit lower than the platform power limit shared by the integrated coprocessor and the discrete coprocessor, and is further configured to keep the sum of a first power provided to the integrated coprocessor and a second power provided to the discrete coprocessor below the reduced platform power limit.
2. The device of claim 1, wherein the power distribution circuit is further configured to: The sum of the first power and the second power is compared with the reduced platform power limit to determine subsequent power allocation conditions.
3. The device of claim 2, wherein the power distribution circuit is further configured to increase the second power provided to the discrete coprocessor in response to the sum being less than the reduced platform power limit.
4. The device as claimed in claim 3, further comprising: At least one sensor configured to measure at least one temperature; and The power distribution circuit is further configured to increase the second power supplied to the discrete coprocessor by an amount determined based on a comparison of the at least one temperature with at least one thermal constraint.
5. The device of claim 4, wherein the power distribution circuit is further configured to: The surface temperature is determined based on the at least one temperature, and The second power increase provided to the discrete coprocessor is determined by a comparison between the surface temperature and the maximum surface temperature determined by the at least one thermal constraint.
6. The device of claim 5, wherein the power distribution circuit is further configured to reduce the second power supplied to the discrete coprocessor in response to the surface temperature exceeding the maximum surface temperature.
7. The device of claim 1, wherein the power distribution circuit is further configured to modify the power supplied to the integrated coprocessor and the discrete coprocessor in response to a change in the characteristics of the workload executed on at least one of the integrated coprocessor and the discrete coprocessor.
8. The device of claim 7, wherein the power distribution circuitry is further configured to transfer power from the integrated coprocessor to the discrete coprocessor in response to a change in workload from core-intensive to graphics-intensive.
9. The device of claim 7, wherein the power distribution circuitry is further configured to transfer power from the discrete coprocessor to the integrated coprocessor in response to a workload shifting from graphics-intensive to core-intensive.
10. A method comprising: A first power is determined to be provided to an integrated coprocessor comprising a first integrated coprocessor core and a second integrated coprocessor core, the integrated coprocessor being configured to generate commands for execution on a discrete coprocessor external to the integrated coprocessor; Determine the second power supplied to the discrete coprocessor; as well as Based on the characteristics of the workloads executed on the integrated coprocessor and the discrete coprocessor, and based on a reduced platform power limit below the platform power limit shared by the integrated coprocessor and the discrete coprocessor, at least one of the first power and the second power is modified; and The sum of the first power and the second power is kept below the reduced platform power limit.
11. The method of claim 10, further comprising: The sum of the first power and the second power is compared with the reduced platform power limit to determine subsequent power allocation conditions.
12. The method of claim 11, further comprising: In response to the sum being less than the reduced platform power limit, the second power provided to the discrete coprocessor is increased.
13. The method of claim 12, further comprising: Measure at least one temperature associated with at least one of the integrated coprocessor and the discrete coprocessor; as well as The second power increase provided to the discrete coprocessor is determined by a comparison of the at least one temperature with at least one thermal constraint.
14. The method of claim 13, further comprising: The surface temperature is determined based on at least one of the temperatures; The surface temperature is compared with the maximum surface temperature determined by the at least one thermal constraint; as well as The second power supplied to the discrete coprocessor is increased by an amount determined based on the comparison.
15. The method of claim 14, further comprising: In response to the surface temperature exceeding the maximum surface temperature, the second power supplied to the discrete coprocessor is reduced.
16. The method of claim 10, further comprising: In response to a change in the characteristics of the workload executed on at least one of the integrated coprocessor and the discrete coprocessor, the first power supplied to the integrated coprocessor and the second power supplied to the discrete coprocessor are modified.
17. The method of claim 16, further comprising: In response to the workload changing from core-intensive to graphics-intensive, power is shifted from the integrated coprocessor to the discrete coprocessor.
18. The method of claim 16, further comprising: In response to the workload changing from graphics-intensive to core-intensive, power is transferred from the discrete coprocessor to the integrated coprocessor.
19. An apparatus comprising: An integrated coprocessor, comprising a central processing unit (CPU) and an integrated graphics processing unit (iGPU), wherein the CPU is configured to generate commands for execution on the iGPU and the discrete coprocessor; and A power distribution circuit is configured to dynamically transfer power between the integrated coprocessor and the discrete coprocessor based on a reduced platform power limit lower than that shared by the integrated coprocessor and the discrete coprocessor, and in response to changes in the characteristics of the workloads executed on the integrated coprocessor and the discrete coprocessor. The power distribution circuit is also configured to keep the sum of the first power provided to the integrated coprocessor and the second power provided to the discrete coprocessor below the reduced platform power limit.
20. The apparatus of claim 19, further comprising: At least one sensor configured to measure at least one temperature, and The power distribution circuit is configured to modify the power supplied to at least one of the integrated coprocessor and the discrete coprocessor in response to a change in the at least one temperature. The power distribution circuit is configured to exchange workload activity requirements and power limit information with the discrete coprocessor.
Citation Information
Patent Citations
Dynamic CPU GPU load balancing using power
CN104106053A
Computing system employing a multi-GPU graphics processing and display subsystem supporting single-GPU non-parallel (multi-threading) and multi-GPU application-division parallel modes of graphics processing operation
US20110169840A1
Changing power limits based on device state
US20160266629A1
