Clock mode and power consumption

[b_r_m_c_cool]

Hi nvidia,

I want to know about 3 contents.

1st,
Is there a most efficient clock mode such as MAX-Q at TX2?
#of course I understand there are 7 mode by setting nvpmodel

2nd,
VDD_CV includes DLA power and PVA power?
If so, when calculating power consumption at deep learning by GPU and DLA,
I should measure VDD_GPU+VDD_CPU+VDD_CV?

3rd,
I measured power consumption at VGG16, VDD_CPU is higher than GPU.
What is the reason for this?

condition:
clock mode : MAXN
weights : FP32
GPU : yes
Framework : tensorRT

result:
VDD_GPU(MAX) : 2162 mW
VDD_CPU(MAX) : 4474 mW
VDD_SoC(MAX) : 3860 mW
VDD_CV(MAX) : 0
VDD_VDDQ(MAX) : 926 mW
VDD_SYS5V(MAX) : 3577
#Both peak and avg CPU is higher than GPU

[Trumany]

1.Max-Q mode provides equivalent performance at full clock modes, no such mode on Xavier.

2.VDD_CV includes DLA and PVA power.

3.The result should be okay, you can check CPU/GPU utilization with ./tegrastat if you want to dig it further.