Jetson watt measures

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
#1

Hey,

We measured the Jetson Xavier input power running in MAXN in two ways:

  1. We measured with an external Clamp meter running on the Jetson power input line

  2. We read the internal Nvidia measures as describe here

As for Nvidia there are 4 parts:

GPU,CPU,SOC,DRR Nvidia measures 12V as an input for 3/4 of them and 5V for the DRR. Therefore the total watt is lower than our external measurement.

Even in maxn mode nvidia internal measures reports ~29-32 watt, comparing the external reports that measure 46-50W.

Any ideas why are we getting those differences?

Thanks

Alon

EDIT
We are following

image
image810×423 46.3 KB

combine all of them on Jetson Xavier with MAXN does not pass 30W, but on external probing we get around 50W

EDIT2
Output of the reading on MAXN
image

1 Like
#3

hello alon2,

could you please also check reported statistics with tegrastats utility.
thanks

#4

Hey JerryChang,
Any reason for that to report something different?

#5

hello alon2,

please check reported statistics with tegrastats utility and share the results.
thanks

#7

Hey @JerryChang I took us a while but here is it:

  1. external reading on the clam we get 45.12W
    2.when reading like the original message we can overall of 27896W, as can be seen in the following image
    image
  2. here is the output of tegrastate at the same time.

image
image1638×39 5.1 KB

#9

hello alon2,

please refer to my results by checked via INA3221 power monitor on Jetson Xavier NX,
for example,
check in_power*_input to gets rail power in milliwatts.

$ sudo -i 
# cd /sys/bus/i2c/drivers/ina3221x/7-0040/iio:device0
# cat in_power0_input
3429
# cat in_power1_input
489
# cat in_power2_input
1224
#10

Hey, the image above contain those value from those value. the last column is the watt .image. In the 2nd column you can see the device name as the rail calls it.

We followed the image|690x360

#11

Note that the mismatch between external measure and those internal as you suggest are the reason for this ticket.

#12

hello alon2,

may I know where did you recap the Table 5-1, Address and Channel Allocation?
Jetson Xavier NX module has one INA3221 power monitor at I2C address 0x40.
thanks

#13

Hey, we are running on
Jetson AGX Xavier, not the NX module.

#14

move this topic to Jetson AGX Xavier.

#15

hello alon2,

could you please also share the connections for your external probing.
thanks

#16

Sure, we are using Fluke 376 to measure to current on the input.

image
image1020×570 32.6 KB

#17

hello alon2,

please refer to Jetson AGX Xavier OEM Product Design Guide,
please check [5.6.2 Power Monitor] session for reference, are you measure the same as the diagram shown?
thanks

#18

Hey,

We measure the output from the transformation. We multiplied the current on this line with voltage and got the watt.

Since we are operating on MAXN, it does make sense that the watt is more than 30Watt. But it does not make sense that the internal measure is less than 30 watts. What do you think?

Current I have two hypothesis:

  1. there is another part on the Jetson that we did not include in the measures.

  2. There is a bug/an error in the internal measurement.

What do you think ?

#20

Hi, the power monitor values of tegrastats only reflect some main parts of power consumption, not for all. Also to get accurate external power value, it needs DAQ device and sw application.

#21

Hey @Trumany, So to sum it, we must use external hardware to read the total power consumption of the Jetson, right?

#22

Yes if you want to get accurate total power consumption.