• State TI Thinks Resolved
  • Locked Locked
  • Replies 5 replies
  • Answers 1 answer
  • Subscribers 98 subscribers
  • Views 797 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

AM5728: OPP settings and power consumption

yf liao
Expert 1080 points
Part Number: AM5728
Other Parts Discussed in Thread: AM5718,

Hi,

  1. Under each OPP level there is a max freq. for each domain, then is there a min freq. under each OPP level?
  2. In comparing with AM572x thermal consideration (SPRAC53–December 2016 ) and AM571x thermal consideration (SPRACI0–October 2018 ), it turns out that AM5728 in single core has lower power consumption but higher junction temperature than AM5718. Could you explain the reason behind?

for example, both Dhrystone, OPP_NOM, 25degC ambient, junction temperature AM5728(single core) vs AM5718:

58degC vs 37.4degC, delta: +20.6degC

power consumpiton:

2092mW vs 2987mW, delta: -895mW  

        3. In another document AM572x power consumption summary(SPRAC79–March 2017) and AM571x power consumption summary(SPRAC78–March 2017), the power consumption result are quite different from the measurements in the report SPRAC53–December 2016 and SPRACI0–October 2018.

For example, on AM5718

under same test condition: idle(MPU @HIGH, GPU @HIGH, DSP@HIGH, IVA@HIGH), 25degC ambient,  the SPRACI0–October 2018 measurement vs SPRAC78–March 2017 measurement:

2295mW vs 1307mW

under minor different test condition: Dhrystone, (MPU@HIGH, MPU @HIGH, GPU @HIGH, DSP@HIGH, IVA@HIGH) in SPRACI0–October 2018 vs Dhrysone,  (MPU@HIGH, GPU @NOM, DSP@NOM IVA@NOM) in SPRAC78–March 2017

4092mW vs 2956mW

It looks to me that the result in SPRAC78–March 2017 and SPRAC79–March 2017 more make sense, while in SPRAC53–December 2016 the AM5728 result is lower than expected, in SPRACI0–October 2018 the AM5718 result is much higher than expected.

Please help to clarify which document best reflect the power consumption of AM5728 and AM5718?

  • 0
    TI__Genius 13567 points

    1) There is not a minimum frequency for a given OPP. However please see the datasheet to understand the impact on Power On Hours (POH) versus OPP.

    2)

    There is a difference between the power measurement method used that makes these documents not comparable. The AM572x GP EVM does not expose all the SoC supplies so these power numbers are only for CORE + MPU + DSP + EVE + GPU. This does not include PHY, PLL, LVCMOS I/O, DDR, or DDR-VTT power. For the AM571x document an internal board was used that allowed the full chip power to be collected. The temperature however is comparable between the two documents as they use the same capture method.

    The other documents SPRAC78 and SPRAC 79 are better for comparison. Here is a disclaimer though: while they are meant to be at room temperature we did not force any constant temperature condition. So temperature may have an effect on the numbers. But this is more "real-world" since processor self-heating is now considered.

  • 0 in reply to Ahmad_Rashed
    Expert 1080 points

    Thanks for your clarification.

    Two more questions:

    1. Why 5728 single core still shows much higher power consumption compared to 5718? e.g. 3115mW vs 1845mW in SPRAC78 and SPRAC 79.
    2. Can we turn off anything in the 5728 (single core) to get it working under lower power consumption?

  • 0 in reply to yf liao
    TI__Genius 13567 points

    1)

    In summary... there are half the number of cores, PCIe PHYs, and EMIF on AM571x. So MPU, GPU, DSP, DDR, and PCIe power should be half compared to AM572x. There is also less L3 / MSMC cache on AM571x so the CORE supply should save power there too.

    In detail... It appears the DSP on the AM572x is powered on with an effect of about 278 mW. The DSP power on AM571x is connected to vdd_gpu, however I know in this SPRAC78 app note the DSP is off so the power is 0 mW. On the MPU, there is 313 mW of difference that I expect is due to the second Cortex A15. Same with GPU, the power is 203 mW on AM572x and half the cores should equal half the power at 113 mW on AM571x (the supply is shared with IVA, so the GPU power is likely lower). On SoC DDR power, this is 375 mW vs 210 mW due to half the memory channels. 1.8V PHY power is also different with 180 mW on AM572x and 70 mW on AM571x (the SDK enables PCIe by default so there is power draw there).

    2) Here is a list of modules that can be turned off by modifying the device tree: BB2D, CAL, DSP, HDMI, IPU, IVA, PCIe, SATA, GPU, USB, VIP

    If all of these are off, I am able to achieve 1.1 W for the whole chip at idle. See the below numbers I captured during some experimentation some time ago.

    In this table, we have the AM5728 with the Processor SDK 4.3 default on the left, and the above listed modules powered off on the right. As mentioned, DSP is off by default on current SDKs so you can see 0 mW for that row. Turning off GPU and IVA will achieve 0 mW on their respective supplies. You can see the effect on USB as well affecting vusb_3v3". And finally you can see the effect of disabling PCIe, SATA, and HDMI on the vdda_1v8_phy supply. 

    AM5728 Default Config mW Lower Power mW Delta
    vdd_core 458 382 76
    vdd_ddr_soc 362 335 26
    vdd_dspeve 0 0 0
    vdd_gpu 244 1 244
    vdd_iva 23 0 22
    vdd_mpu 254 238 15
    vddshv_3v3 73 70 4
    vdds18v 95 68 26
    vdda_1v8_phy 175 33 142
    vdda_1v8_pll 47 40 7
    vusb_3v3 5 0 5
    Total (excl. DDR) 1737 1169 589

    In this table, we have the AM5718 with the Processor SDK 4.3 default on the left and the above mentioned modules powered off on the right. The vdd_gpu+dsp+iva supply is near 0 mW and the PHY supply power is reduced due to disabling PCIe, SATA and HDMI. 

    AM5718 Default Config mW Lower Power mW Delta
    vdd_core 494 411 83
    vdd_ddr_soc 201 188 14
    vdd_gpu+dsp+iva 151 5 147
          0
          0
    vdd_mpu 79 78 1
    vddshv_3v3 159 158 1
    vdds18v 100 79 21
    vdda_1v8_phy 195 44 151
    vdda_1v8_pll 63 58 5
    vusb_3v3 7 0 7
    Total 1451 1020 431
  • 0 in reply to Ahmad_Rashed
    Expert 1080 points

    Hi, just to double confirm, for those disabled PHY, their power rail does not necessarily have to be seperated from others that is in use? for example, in our design, HDMI, USB3, PCIE phy power are tied together, we still can disable PCIE while keep HDMI, USB3 in use?

  • 0 in reply to yf liao
    TI__Genius 13567 points

    yf liao said:

    In our design, HDMI, USB3, PCIE phy power are tied together, we still can disable PCIE while keep HDMI, USB3 in use?

    Yes, the PCIE PHY will not use any power when disabled but the 1.8V must be connected.