Just to clarify - the new values you see in the latest HWiNFO Beta are not obtained via I2C - they use GPU's own telemetry methods.
The VRMs are accessible, but the problem is that on the Fiji any attempt to access the GPU I2C bus can result in a system crash (black screen + fan @ max). I have discussed this with AMD and followed their recommendations how to access the I2C safely. That seemed to improve the situation, but still is not stable enough. It seems that on the (non-X) Fury the crash is less likely to occur, however on the Fury X I can get the system down within a few seconds when using GPU I2C. My understanding is that the GPU (most probably its SMC or drivers) are heavily communicating using the I2C and even when trying to synchronize with them still can lead to a crash.
So while HWiNFO32/64 v5.20 had GPU I2C support enabled on Fiji (might need to do "Reset GPU I2C Cache" first), I have decided to disable this support in the default mode in later versions. So all later versions will only attempt to access the GPU I2C when you enable the "GPU I2C via ADL" option. But I don't recommend this because of the mentioned problems.
I'm not sure if AMD will fix this (nor whether it's possible to fix this at all), so I recommend to rely on current alternate values, which seem to be quite reliable and stable. Though they won't be able to offer all the details as read directly from the GPU VRMs.
The new Beta can now safely report:
- GPU VR VDDC Temperature
- GPU VRM VDD Temperature
- GPU Liquid Temperature (Fury X only)
- GPU HBM Temperature - doesn't seem to be supported by current GPUs/drivers
- GPU Core Voltage (VDDC)
- GPU Memory Voltage (MVDDC)
- GPU Core Current [Amps]
- GPU Memory Current [Amps]
- GPU Core Power [Watts]
- GPU Memory Power [Watts]