• Menu
  • Product
  • Email
  • PDF
  • Order now
  • AM574x thermal considerations

    • SPRACL7 March   2019 AM5746 , AM5748 , AM5749

       

  • CONTENTS
  • SEARCH
  • AM574x thermal considerations
  1.   AM574x thermal considerations
    1.     Trademarks
    2. 1 Overview
    3. 2 Important Notes
    4. 3 Test Overview
      1. 3.1 OS Idle
      2. 3.2 Dhrystone
      3. 3.3 Temperature Measurement
      4. 3.4 OPP Definitions
    5. 4 Data and Results
      1. 4.1 OPP Settings and Linux Thermal Framework
      2. 4.2 Power and Thermal Chamber Measurements
      3. 4.3 OS Idle at OPP_NOM
      4. 4.4 Dhrystone 1 Core at OPP_NOM
      5. 4.5 Dhrystone 2 Core at OPP_NOM
      6. 4.6 Dhrystone 1 Core at OPP_HIGH
      7. 4.7 Dhrystone 2 Core at OPP_HIGH
    6. 5 References
  2. IMPORTANT NOTICE
search No matches found.
  • Full reading width
    • Full reading width
    • Comfortable reading width
    • Expanded reading width
  • Card for each section
  • Card with all content

 

APPLICATION NOTE

AM574x thermal considerations

AM574x thermal considerations

This application report discusses the thermal performance of the Sitara™ AM574x series processors. The data presented demonstrates the effects of different thermal management strategies in terms of processor junction temperature and power consumption across MPU loading and ambient temperature.

Trademarks

Sitara is a registered trademark of Texas Instruments.

Arm, Cortex are registered trademarks of Arm Limited.

Linux is a registered trademark of Linus Torvalds.

All other trademarks are the property of their respective owners.

1 Overview

In this experiment, an internal AM574x board is used to gather thermal data with different processor loading and ambient temperature. Ambient temperature is controlled with a programmable environmental chamber.

The collected data can be utilized to correlate the thermal performance of the processor and power consumption at a given processor load and junction temperature, based on ambient temperature and thermal management.

Tests were repeated with the following thermal management:

  • Bare package (no heatsink)
  • Low-cost heatsink (31-mm x 31-mm x 19.5-mm, #ATS-54310R-C1-R0)
  • Low-cost heatsink + Fan (5 V, 9500 RPM, 4.9 CFM, #MC30060V1-000U-A99)

2 Important Notes

The environmental chamber used to collect this data circulates air internally to maintain homogeneous internal temperature, and does not accurately simulate the environment on the bench or end product. This is important to consider in passive cooling applications where air circulation can significantly impact PCB, package, and heatsink power dissipation efficiency.

The data presented in this test is gathered with a typical device, representing nominal silicon process and leakage. Thermal performance and power consumption can vary significantly due to process variation. Extra margin must be designed in to account for worst case process variation (leakage).

3 Test Overview

The following CPU loading schemes are characterized with the internal AM574x board for this report.

3.1 OS Idle

The AM574x processor is idling after booting the out-of-box configuration of Processor SDK Linux® v05.02.00. No display was connected to the AM574x board. MPU, GPU, and IVA cores are powered but automatically clock gated while the DSP and IPU cores are both power and clock gated.

3.2 Dhrystone

Dhrystone is a single-threaded benchmark, capable of utilizing approximately 100% of one Arm®Cortex®-A15 core. Dhyrstone is included in the TI Processor SDK. Tests are conducted with the A15 running at 1.0 GHz (OPP_NOM) and 1.5 GHz (OPP_HIGH).

3.3 Temperature Measurement

Reported temperature data is measured by on-die sensors to the approximate actual junction temperature. Temperature for each use-case is measured after soaking for five minutes. Under lab conditions, it is determined a five minute period allows the processor to reach stable temperature.

The TI Processor SDK provides Linux drivers for these sensors, and can be queried from the command-line. For example:

# cat /sys/class/thermal/thermal_zone0/temp 71800

3.4 OPP Definitions

Operating performance points (OPP) levels define a max frequency per fixed voltage level in each voltage domain. Table 1 lists the frequency of each subsystem per OPP for the AM574x processor.

Dynamic Voltage Frequency Scaling (DVFS) refers to a software technique where the system-on-chip (SoC) supplies with AVS support are changed from one OPP level (voltage and frequency pair) to another to either adapt to a changing work-load, or to avoid device operation outside of desired temperature bounds.

This SoC only supports DVFS on the MPU domain. For DSP and GPU domains, the OPP levels must be set during boot by the initial bootloader. Ensure that the selected OPP level meets the needs of the application and all thermal testing is conducted at the desired OPP level.

Table 1. Supported OPP vs Max Frequency

Voltage Domain Clock Domain OPP_NOM OPP_OD OPP_HIGH
Maximum Frequency (MHz) Maximum Frequency (MHz) Maximum Frequency (MHz)
VD_MPU MPU_CLK 1000 1176 1500
VD_DSP DSP_CLK 600 700 750
EVE_FCLK 535 650 650
VD_IVA IVA_GCLK 388.3 430 532
VD_GPU GPU_CLK 425.6 500 532
VD_CORE DDR3 / DDR3L 667 (DDR3-1333) N/A N/A
CORE_IPUx_CLK 212.8 N/A N/A
L3_CLK 266 N/A N/A
VD_RTC RTC_FCLK 0.034 N/A N/A

4 Data and Results

This section contains the raw data and graphs of the test experiments described above. All data is gathered running the latest Linux Processor SDK. All tests are conducted without an external display installed.

4.1 OPP Settings and Linux Thermal Framework

Tests are conducted with Processor SDK 04.02.00 at following OPP levels:

Default OPP Levels
MPU GPU DSP IVA
NOM HIGH HIGH HIGH

The MPU domain OPP defaults to NOM when idle and increases to HIGH when under load. OPP levels for DSP and IVA cores can be changed but that must be done by editing the U-Boot defconfig file with the desired OPP level and recompiling.

Applicable to the AM574x, the Linux kernel on this device uses the CPUFreq driver to support multiple OPPs for the MPU domain and dynamically changes between them. As such, a desired maximum frequency must be set if you are seeking power consumption at a frequency lower than the maximum.

Additionally, the Linux thermal framework needs to be disabled, otherwise, the maximum frequency is reduced as the MPU heats up to prevent thermal shutdown. This must only be done for data gathering purposes when the junction temperature exceeds the levels defined in the device tree and is not recommended for a production system.

 

Texas Instruments

© Copyright 1995-2025 Texas Instruments Incorporated. All rights reserved.
Submit documentation feedback | IMPORTANT NOTICE | Trademarks | Privacy policy | Cookie policy | Terms of use | Terms of sale