16-Pin HVPWR Power consumption higher than GPU

[Takezo]

It is normal to have the 16-pin power consumption higher than the gpu itself (5090)?

Sometimes it even exceed 600W and goes all the way up to 650W while the GPU is stuck at 580 max

i'm risking cable melting?

 

[KevinKaesmann]

Only NVIDIA knows...

Comparing values of OCCT, HWINFO64, GPU-Z while running OCCT GPU Test (adaptive and standard)

...on nearly idle:

• OCCT / HWINFO "GPU Power" and GPU-Z "Board Power" seems nearly identical (slightly different sampling intervals !)
• GPU-Z Board Power Draw seems the sum of "PCIe Slot power" and "16-Pin Power" (8.7 + 26.1 = 34.8 Watt)
• HWINFO "GPU PCIe +12V Input Power" and "GPU 16-pin HVPWR" isn't exactly "GPU Power" (30.7 + 9.377 = 40.077 Watt vs. 35.803 Watt "GPU Power")

...now on 70 Watt:

• OCCT (GPU adaptive, shortly after small increase) 71.94 Watt fits to HWINFO 65.706 Watt / GPU-Z 65.5 Watt
• GPU-Z PCIe Slot Power + 16-Pin Power = Board Power Draw (12.7 + 52.7 = 65.4 Watt)
• HWINFO GPU PCIe +12V + GPU 16-pin HVPWR (13.169 + 55.348 = 68.517 Watt vs. GPU Power 65.706 Watt)

...now on 200 Watt:

• OCCT "GPU Power" 194.262 Watt fits to HWINFO "GPU Power" 194.262 Watt fitts to GPU-Z 196.9 Watt "Board Power Draw"

...now on 240 Watt and OCCT GPU Standard = constant load:

• OCCT "GPU Power" 241.26 Watt, HWINFO "GPU Power" 241.365 Watt and GPU-Z "Board Power Draw" 240.8 Watt.+

My conclusion:
Loading the GPU constantly without "spikes", the "GPU Power" values show reasonable values on all three programs.
But - as I've seen the same spikes in the past, in my opinion, the NVIDIA 16-pin HVPWR Power sensors delivers very short term maximums
while "GPU Power" seems a more "smooth" value.