GPU Vcore & other readings inaccurate with RX 6800 & RX 6900 XT

[cs9k]

Hello! I searched around but did not find anyone that has reported this bug yet.

HWINFO64 is not correctly reading voltages from AMD GPU's in two systems I own.

Files are from the RX 6800 system. Driver version is Adrenalin 21.3.2 on both systems.

Max voltage for the RX 6800 is 1025mV, but is being read at mid 900mV's. Likewise, the RX 6900 XT max voltage is 1175mV, but is being read out at a max of 1117mV. Values were checked against Adrenalin control panel and GPUZ 2.38.0.

I suspect the voltage readings for GPU Memory Voltage (VDDCI_MEM) and GPU SoC Voltage (VDDCR_SOC) could be incorrect as well.

 

[Martin]

Which exact values in HWiNFO are you checking, can you post a screenshot? Best would be a side-by-side screenshot with Adrenalin.

 

[cs9k]

The only value I can prove that is not consistent is GPU Core Voltage (VDDCR_GFX), as seen in the screencap. Unigine Heaven was running on my main screen at the time of the screenshot.

I suspect the voltage readings for GPU Memory Voltage (VDDCI_MEM) and GPU SoC Voltage (VDDCR_SOC) could be incorrect as well, but I have no way of proving that, other than on my RX 6900 XT system, those two values aren't what I believe they should be.

 

[Martin]

All values HWiNFO is showing are correct.
The GPU voltage shown in AMD Radeon Software and GPU-Z is only the last voltage requested. What HWiNFO is showing is the current voltage provided. So similar to difference between CPU VID and Vcore.

 

[cs9k]

Very interesting! I did some more messing around with settings and it would seem that the GPU Core Voltage reading (limit) that I'm seeing is some artifact of the firmware or drivers limiting voltage or current to the core, for reasons I can't currently explain. Under partial loads, the core voltage will indeed display peaks at the 1175mV limit in HWINFO64, but under full load it sails right up to 1115mV and stops, which is preventing me from pushing the card's core clock further.

Can't blame AMD if it is indeed a firmware/driver limit, just annoying to run into with my GPU wearing a water block. We can close this ticket/issue.

Thanks for your help and insight!