No idea on what is going on
The Stilt stated output loop 2 is not used in ref PCB RX 480 when the whole powergate thing was going on. Therefore using register 8E shouldn't have an effect. Due to it not being used when The Stilt released ROMs for RX 480 powergate issue his ROMs only had VMAX for loop 1 and not 2, his Hawaii ROMs always had VMAX for both loops as they were both used.
Perhaps as Eliovp/Heliox is using a Nitro something differs on it. I am not experienced or knowledgeable enough to comment from viewing say a high res Nitro PCB image to say what is connected which way in this case.
Without a voltage control chip that has a data interface adjusting a voltage via ROM/SW would have no effect. For example there is VDDCI value in Fiji PowerPlay but it has no negative or positive effect on card's operation when adjusted, as there is no programmable voltage control chip for VDDCI voltage on PCB.
To know whats going on when an owner is adjusting what is labeled in Wattman as memory voltage we need information where someone has taken a voltage reading using a digital multimeter say for VDDCI, has anyone posted this information?
I no way refute the posts where members have stated when they change memory voltage in Wattman it has an effect on what RAM clock they can achieve/sustain, but this is not clear proof it is VDDCI or MVDD IMO. Perhaps Polaris has SMC like Fiji and the GPU on-die controller is adjusting a voltage explaining why members see an effect, I don't know
If MSI AB is now supporting voltage control on Polaris it would be interesting to see the MSI AB info screen, this would also confirm if Polaris is using i2c or SMC messaging for that tool IMO.
On Hawaii/Grenada cards when you click the "i" in MSI AB window you would see a screenie like this:-Warning: Spoiler! (Click to show)
The MSI 290X Lightning had IR3567B & IR3570B, when you click the "i" in MSI AB window you would see a screenie like this:-Warning: Spoiler! (Click to show)
On Fiji you would see this :-Warning: Spoiler! (Click to show)
From what I understand/tested on Fiji, MSI AB,etc message SMC via driver, then SMC sends command to IR3567B, and I then see the register for voltage offset change in i2c dump. As SMC is always keeping i2c line between it's self and IR3567B busy, i2c dump takes a lot longer than Hawaii using MSI AB (so far AIDA64 lacks support for this on Fiji). Even though Unwinder gave some pointers on SMC messaging on Guru3D/OCN I found no info on HBM voltage adjustment SMC message command (ie output loop 2). MSI AB not only supports i2c operations via command line but also SMC messaging, there is a post on Guru3D by Unwinder (author of MSI AB).Edited by gupsterg - 9/18/16 at 8:27am