DRA746: Software thermal simulation

Hi, Dharshan BM,

The expert's answer is bellow:

From a S/W perspective, stressing the system as much as possible, will result in higher temperatures, some items that can be focused on:

1) Ensure clocks frequencies are at the highest values (see OPP_HIGH in DRA746 data manual)
2) S/W utility to maximize CPU usage
3) S/W utility to maximize GPU usage
4) S/W utilities to maximize DSP/IPU usage
5) Maximize memory usage if (2) (3) and (4) are not already achieving this.

S/W utilities available, and clock frequency management options, will be dependent on the OS version that is under test. If you could please provide OS version, a more detailed response can be provided.


Regards,
Mariya

Hi Dharshan,

Thanks for the OS information. Regarding the provided list of items to try:

(1) To ensure clock frequencies are set at OPP_HIGH, the IPL code can be rebuilt with the *_OPP_HIGH compile flags set to 1. After bootup with the newly generated MLO, if the QNX DVFS framework is running then ensure it is locking MPU to OPP_HIGH.

(2) To maximize CPU usage, recommend coding a simple utility that consists of a tight loop, that does some operation on a volatile variable. A configurable delay within the loop, would allow some control on how much CPU is consumed.

(3) To maximize GPU usage, multiple instances of the gles2-gears, or gles1-gears can be run. Setting the command line option of interval=0, will maximize FPS.

(4) To maximize IPU/DSP usage, would depend on the images that are available to be run on those cores. If not available, alternatively a video decode could be done to increase system stress.

(5) With (1) to (4) running concurrently no additional memory B/W consumption should be required.

Regards,

kb