Yes, the polling rate affects it but it will be still too large for such a long period.
Try to disable monitoring of the PresentMon sensor if that will solve the problem.
Not sure what exactly do you mean without seeing a screenshot but if you have a GIGABYTE board then it's most likely the AORUS theme used by HWiNFO. This can easily be disabled in main settings.
Close HWiNFO.
Run "regedit" tool
Navigate to: HKEY_CURRENT_USER\Software\HWiNFO64\Sensors\GpuI2cCache
Delete the entire key.
If that won't help, please attach the HWiNFO Debug File.
If voltage offsets are not shown, then most likely because Windows HVCI (Core isolation/Memory integrity) is blocking access and you will need to disable it to see them.
I see the GPU load fluctuates a lot, so the temperature should be as well. In such case when each app measures the value at a different time point, it could be different. Can you compare the values at constant GPU load?
Also, another reason could be a bug in NVIDIA drivers when temperature...
HWiNFO reads the actual timings and clock straight from the memory controller.
So your RAM is really running only the standard JEDEC timing and XMP wasn't properly applied by BIOS.