About the power consumption of TMS320C6678

Jason,

I'm assuming this is the total board power consumption, is that correct?

It's difficult to say for certain what all you're getting w/ such a measurement but I'll try to give you some high level information about power consumption on devices.

As we move to smaller and smaller process nodes, more and more power consumption is due to leakage current of the devices.  The leakage current is exponential w/ the temperature of the device and also the square of the voltage level.  This is a major reason for Smart Reflex implementation.

The actual power is going to vary from device to device, but we screen to a max limit of the power spreadsheet (should be available at the end of August.)

That said, the 9.6W could have been measured when the device was first turned on, and hadn't heated up yet to higher temps (and thus higher power consumption.)  And that's true for other devices on the board as well.

For the Demo vs 1 core FFT vs 8 core FFT all be 12.3W.  I'd assume it was measured when the device temp stabilized.   So why would it all show the same power consumption.  Depending on how you measured it, where you measured it and how long the code is running your observations may be different.

If you're measuring it at the power rails using a current loop and an Oscilloscope you're going to see immediate power on the scope with fine resolution over time.  If you're using a power meter at the wall w/ the power supply plugged in, you're going to get periodic samples and much of the power consumption could be buffered in the supplies (both the power brick and the on board supplies.)  And if you're doing single shot runs of the code, you're most likely missing the actual power consumed when running the executable.  You want to make sure these are running in an infinite loop w/ little to no down time in between execution.

I hope this helps a bit.  Let me know what the conditions are of your measurements and we can discuss in more details.

Best Regards,

Chad

Hi Chad:

Thank you for your timely reply.

When we test the power consumption of the Demo board, these are running in an infinite loop and we use the power meter to make the measure.

The temperature is stablized. When the power consumption rise to 12.3W , if we push the reset button of the board, it will return to 9.6w untill we load

program to tms320c6678.

I think there should be some power consumption rising due to the FFT, but there isn't.

When cores doing FFT in an infinite loop, switching current should be much larger than no FFT, is it right?

I'd expect it to be larger, but not much larger - maybe 100-200mW/core used.

When you say power meter, are you referring to one plugged into the wall and the power supply to the board plugged into that?  This would be a very sub-optimal measurement.  You will have included everything on the board, the power supply loss, any other activity going on, etc.

Also, the 9.6 to 12.3W delta is more than the DSP, as more things are held in rest than just the DSP when the board reset is applied.

Also, for the FFT, how big of an FFT are you doing, how many cycles between exiting and re-entering the loop?  Where's the memory stored?  Are you getting the expected throughput performance or are you effectively starving for data so the actual instruction execution is much slower than if the data was resident in L1D or L2 w/ L1D cache.

It's hard for me to say at this 10K Foot level what all is going on.

Best Regards,

Chad